announced the judgment of the Court and delivered an opinion, in which The Chief Justice and Justice Kennedy join.
This is the latest in a series of appeals involving racial gerrymandering challenges to state redistricting efforts in the wake of the 1990 census. See Shaw v. Hunt, ante, p. 899 (Shaw II); United States v. Hays, 515 U. S. 787 (1995); Miller v. Johnson, 515 U. S. 900 (1995); Shaw v. Reno, 509 U. S. 630 (1993) (Shaw I). That census revealed a population in*957crease, largely in urban minority populations, that entitled Texas to three additional congressional seats. In response, and with a view to complying with the Voting Rights Act of 1965 (VRA), 79 Stat. 437, as amended, 42 U. S. C. § 1973 et seq., the Texas Legislature promulgated a redistricting plan that, among other things, created District 30, a new majority-African-American district in Dallas County; created District 29, a new majority-Hispanic district in and around Houston in Harris County; and reconfigured District 18, which is adjacent to District 29, to make it a majority-African-American district. The Department of Justice pre-cleared that plan under VRA § 5 in 1991, and it was used in the 1992 congressional elections.
The plaintiffs, six Texas voters, challenged the plan, alleging that 24 of Texas’ 30 congressional districts constitute racial gerrymanders in violation of the Fourteenth Amendment. The three-judge United States District Court for the Southern District of Texas held Districts 18, 29, and 30 unconstitutional. Vera v. Richards, 861 F. Supp. 1304 (1994). The Governor of Texas, private intervenors, and the United States (as intervenor) now appeal. We noted probable jurisdiction. 515 U. S. 1172 (1995). Finding that, under this Court’s decisions in Shaw I and Miller, the district lines at issue are subject to strict scrutiny, and that they are not narrowly tailored to serve a compelling state interest, we affirm.
I
As a preliminary matter, the State and private appellants contest the plaintiffs’ standing to challenge these districts. Plaintiff Chen resides in Texas congressional District 25, and has not alleged any specific facts showing that he personally has been subjected to any racial classification. Under our decision in Hays, he lacks standing. See Hays, supra, at 744-745. But plaintiffs Blum and Powers are residents of District 18, plaintiffs Thomas and Vera are residents of District 29, and plaintiff Orcutt is a resident of District 30. We *958stated in Hays that “[w]here a plaintiff resides in a racially gerrymandered district, . . . the plaintiff has been denied equal treatment because of the legislature’s reliance on racial criteria, and therefore has standing to challenge the legislature’s action.” Ibid.; accord, Miller, supra, at 910-911. Under this rule, these plaintiffs have standing to challenge Districts 18, 29, and 30.
II
We must now determine whether those districts are subject to strict scrutiny. Our precedents have used a variety of formulations to describe the threshold for the application of strict scrutiny. Strict scrutiny applies where “redistricting legislation ... is so extremely irregular on its face that it rationally can be viewed only as an effort to segregate the races for purposes of voting, without regard for traditional districting principles,” Shaw I, supra, at 642, or where “race for its own sake, and not other districting principles, was the legislature’s dominant and controlling rationale in drawing its district lines,” Miller, 515 U. S., at 913, and “the legislature subordinated traditional race-neutral districting principles ... to racial considerations,” id., at 916. See also id., at 928 (O’Connor, J., concurring) (strict scrutiny only applies where “the State has relied on race in substantial disregard of customary and traditional districting practices”).
Strict scrutiny does not apply merely because redistricting is performed with consciousness of race. See Shaw I, supra, at 646. Nor does it apply to all cases of intentional creation of majority-minority districts. See DeWitt v. Wilson, 856 F. Supp. 1409 (ED Cal. 1994) (strict scrutiny did not apply to an intentionally created compact majority-minority district), summarily aff’d, 515 U. S. 1170 (1995); cf. Shaw I, supra, at 649 (reserving this question). Electoral district lines are “facially race neutral,” so a more searching inquiry is necessary before strict scrutiny can be found applicable in redistricting cases than in cases of “classifications based explicitly on race.” See Adarand Constructors, Inc. v. Peña, 515 U. S. *959200, 213 (1995); cf. post, at 999-1000, 1002-1003 (THOMAS, J., concurring in judgment) (assimilating our redistricting cases to Adarand). For strict scrutiny to apply, the plaintiffs must prove that other, legitimate districting principles were “subordinated” to race. Miller, 515 U. S., at 916. By that, we mean that race must be “the predominant factor motivating the legislature’s [redistricting] decision.” Ibid, (emphasis added). We thus differ from Justice Thomas, who would apparently hold that it suffices that racial considerations be a motivation for the drawing of a majority-minority district. See post, at 1002.
The present suit is a mixed motive suit. The appellants concede that one of Texas’ goals in creating the three districts at issue was to produce majority-minority districts, but they also cite evidence that other goals, particularly incumbency protection (including protection of “functional incumbents,” i. e., sitting members of the Texas Legislature who had declared an intention to run for open congressional seats), also played a role in the drawing of the district lines. The record does not reflect a history of “ ‘purely race-based’ ” districting revisions. Cf. Miller, supra, at 918 (emphasis added). A careful review is, therefore, necessary to determine whether these districts are subject to strict scrutiny. But review of the District Court’s findings of primary fact and the record convinces us that the District Court’s determination that race was the “predominant factor” in the drawing of each of the districts must be sustained.
We begin with general findings and evidence regarding the redistricting plan’s respect for traditional districting principles, the legislators’ expressed motivations, and the methods used in the redistricting process. The District Court began its analysis by rejecting the factual basis for appellants’ claim that Texas’ challenged “districts cannot be unconstitutionally bizarre in shape because Texas does not have and never has used traditional redistricting principles such as natural geographical boundaries, contiguity, compactness, *960and conformity to political subdivisions.” 861 F. Supp., at 1333. The court instead found that “generally, Texas has not intentionally disregarded traditional districting criteria,” and that only one pre-1991 congressional district in Texas was comparable in its irregularity and noncompactness to the three challenged districts. Id., at 1334. The court also noted that “compactness as measured by an ‘eyeball’ approach was much less important,” id., at 1313, n. 9, in the 1991 plan, App. 144, than in its predecessor, the 1980 Texas congressional districting plan, id., at 138, and that districts were especially irregular in shape in the Dallas and Harris County areas where the challenged districts are located, see 861 F. Supp., at 1313, n. 9.
These findings comport with the conclusions of an instructive study that attempted to determine the relative compactness of districts nationwide in objective, numerical terms. That study gave Texas’ 1980 districting plan a roughly average score for the compactness and regularity of its district shapes, but ranked its 1991 plan among the worst in the Nation. See Pildes & Niemi, Expressive Harms, “Bizarre Districts,” and Voting Rights: Evaluating Election-District Appearances After Shaw v. Reno, 92 Mich. L. Rev. 483, 571-573, table 6 (1993). The same study ranked Districts 18, 29, and 30 among the 28 least regular congressional districts nationwide. See id., at 565, table 3. Our own review gives us no reason to disagree with the District Court that the districts at issue “have no integrity in terms of traditional, neutral redistricting criteria,” 861 F. Supp., at 1339.
The District Court also found substantial direct evidence of the legislature’s racial motivations. The State’s submission to the Department of Justice for preclearance under VRA § 5 reports a consensus within the legislature that the three new congressional districts
“ ‘should be configured in such a way as to allow members of racial, ethnic, and language minorities to elect Congressional representatives. Accordingly, the three *961new districts include a predominantly black district drawn in the Dallas County area [District 30] and predominantly Hispanic districts in the Harris County area [District 29] and in the South Texas region. In addition to creating the three new minority districts, the proposed Congressional redistricting plan increases the black voting strength of the current District 18 (Harris County) by increasing the population to assure that the black community may continue to elect a candidate of its choice.’” Id., at 1315 (quoting Narrative of Voting Rights Act Considerations in Affected Districts, reprinted in App. 104-105).
The appellants also conceded in this litigation that the three districts at issue “were created for the purpose of enhancing the opportunity of minority voters to elect minority representatives to Congress.” 861 F. Supp., at 1337. And testimony of individual state officials confirmed that the decision to create the districts now challenged as majority-minority districts was made at the outset of the process and never seriously questioned.
The means that Texas used to make its redistricting decisions provides further evidence of the importance of race. The primary tool used in drawing district lines was a computer program called “REDAPPL.” REDAPPL permitted redistricters to manipulate district lines on computer maps, on which racial and other socioeconomic data were superimposed. At each change in configuration of the district lines being drafted, REDAPPL displayed updated racial composition statistics for the district as drawn. REDAPPL contained racial data at the block-by-block level, whereas other data, such as party registration and past voting statistics, were only available at the level of voter tabulation districts (which approximate election precincts). The availability and use of block-by-block racial data was unprecedented; before the 1990 census, data were not broken down beyond the census tract level. See App. 123. By providing uniquely *962detailed racial data, REDAPPL enabled districters to make more intricate refinements on the basis of race than on the basis of other demographic information. The District Court found that the districters availed themselves fully of that opportunity:
“In numerous instances, the correlation between race and district boundaries is nearly perfect. . . . The borders of Districts 18, 29, and 30 change from block to block, from one side of the street to the other, and traverse streets, bodies of water, and commercially developed areas in seemingly arbitrary fashion until one realizes that those corridors connect minority populations.” 861 F. Supp., at 1336.
These findings — that the State substantially neglected traditional districting criteria such as compactness, that it was committed from the outset to creating majority-minority districts, and that it manipulated district lines to exploit unprecedentedly detailed racial data — together weigh in favor of the application of strict scrutiny. We do not hold that any one of these factors is independently sufficient to require strict scrutiny. The Constitution does not mandate regularity of district .shape, see Shaw I, 509 U. S., at 647, and the neglect of traditional districting criteria is merely necessary, not sufficient. For strict scrutiny to apply, traditional dis-tricting criteria must be subordinated to race. Miller, 515 U. S., at 916. Nor, as we have emphasized, is the decision to create a majority-minority district objectionable in and of itself. The direct evidence of that decision is not, as Justice Stevens suggests, post, at 1024, “the real key” to our decision; it is merely one of several essential ingredients. Nor do we “condemn state legislation merely because it was based on accurate information.” Post, at 1031, n. 28. The use of sophisticated technology and detailed information in the drawing of majority-minority districts is no more objectionable than it is in the drawing of majority-majority dis*963tricts. But, as the District Court explained, the direct evidence of racial considerations, coupled with the fact that the computer program used was significantly more sophisticated with respect to race than with respect to other demographic data, provides substantial evidence that it was race that led to the neglect of traditional districting criteria here. We must therefore consider what role other factors played in order to determine whether race predominated.
Several factors other than race were at work in the drawing of the districts. Traditional districting criteria were not entirely neglected: Districts 18 and 29 maintain the integrity of county lines; each of the three districts takes its character from a principal city and the surrounding urban area; and none of the districts is as widely dispersed as the North Carolina district held unconstitutional in Shaw II, ante, p. 899. (These characteristics are, however, unremarkable in the context of large, densely populated urban counties.) More significantly, the District Court found that incumbency protection influenced the redistricting plan to an unprecedented extent:
“[A]s enacted in Texas in 1991, many incumbent protection boundaries sabotaged traditional redistricting principles as they routinely divided counties, cities, neighborhoods, and regions. For the sake of maintaining or winning seats in the House of Representatives, Congressmen or would-be Congressmen shed hostile groups and potential opponents by fencing them out of their districts. The Legislature obligingly carved out districts of apparent supporters of incumbents, as suggested by the incumbents, and then added appendages to connect their residences to those districts. The final result seems not one in which the people select their representatives, but in which the representatives have selected the people.” 861 F. Supp., at 1384 (citations and footnotes omitted).
*964See also id., at. 1317-1318 (describing specific evidence of incumbency protection efforts statewide). This finding receives inferential support from the fact that all but one of Texas’ 27 incumbents won in the 1992 elections. See id., at 1318. And the appellants point to evidence that in many eases, race correlates strongly with manifestations of community of interest (for example, shared broadcast and print media, public transport infrastructure, and institutions such as schools and churches) and with the political data that are vital to incumbency protection efforts, raising the possibility that correlations between racial demographics and district lines may be explicable in terms of nonracial motivations. For example, a finding by a district court that district lines were drawn in part on the basis of evidence (other than racial data) of where communities of interest existed might weaken a plaintiff’s claim that race predominated in the drawing of district lines. Cf. post, at 1049 (Souter, J., dissenting) (recognizing the legitimate role of communities of interest in our system of representative democracy).
Strict scrutiny would not be appropriate if race-neutral, traditional districting considerations predominated over racial ones. We have not subjected political gerrymandering to strict scrutiny. See Davis v. Bandemer, 478 U. S. 109, 132 (1986) (White, J., plurality opinion) (“[UJnconstitutional discrimination occurs only when the electoral system is arranged in a manner that will consistently degrade a voter’s or a group of voters’ influence on the political process as a whole”); id., at 147 (O’Connor, J., concurring in judgment) (“[PJurely political gerrymandering claims” are not justicia-ble). And we have recognized incumbency protection, at least in the limited form of “avoiding contests between incumbent[sj,” as a legitimate state goal. See Karcher v. Daggett, 462 U. S. 725, 740 (1983); White v. Weiser, 412 U. S. 783, 797 (1973); Burns v. Richardson, 384 U. S. 73, 89, n. 16 (1966); cf. Gaffney v. Cummings, 412 U. S. 735, 751-754, and 752, n. 18 (1973) (State may draw irregular district lines in order *965to allocate seats proportionately to major political parties). Because it is clear that race was not the only factor that motivated the legislature to draw irregular district lines, we must scrutinize each challenged district to determine whether the District Court’s conclusion that race predominated over legitimate districting considerations, including incumbency, can be sustained.
A
The population of District 30 is 50% African-American and 17.1% Hispanic. Fifty percent of the district’s population is located in a compact, albeit irregularly shaped, core in south Dallas, which is 69% African-American. But the remainder of the district consists of narrow and bizarrely shaped tentacles — the State identifies seven “segments” — extending primarily to the north and west. See App. 335; see also M. Barone & G. Ujifusa, Almanac of American Politics 1996, p. 1277 (1995) (describing the district). Over 98% of the district’s population is within Dallas County, see App. 118, but it crosses two county lines at its western and northern extremities. Its western excursion into Tarrant County grabs a small community that is 61.9% African-American, id., at 331; its northern excursion into Collin County occupies a hook-like shape mapping exactly onto the only area in the southern half of that county with a combined African-American and Hispanic percentage population in excess of 50%, id., at 153. The District Court’s description of the district as a whole bears repeating:
“The district sprawls throughout Dallas County, deliberately excludes the wealthy white neighborhoods of Highland Park and University Park and extends fingers into Collin County, which include the outermost suburbs of Dallas. In Collin County, the district picks up a small African-American neighborhood. The district extends into Tarrant County only to pick up a small border area with a high African-American concentration. It *966also reaches out to claim Hamilton Park, an affluent African-American neighborhood surrounded by whites. Part of the district runs along Trinity River bottom, using it to connect dispersed minority population. Numerous [voter tabulation districts] were split in order to achieve the population mix required for the district.
“. . . It is at least 25 miles wide and 30 miles long.” 861 F. Supp., at 1337-1338.
See also Appendix A to this opinion (outline of District 30).
Appellants do not deny that District 30 shows substantial disregard for the traditional districting principles of compactness and regularity, or that the redistricters pursued unwaveringly the objective of creating a majority-African-American district. But they argue that its bizarre shape is explained by efforts to unite communities of interest in a single district and, especially, to protect incumbents.
Appellants highlight the facts that the district has a consistently urban character and has common media sources throughout, and that its tentacles include several major transportation lines into the city of Dallas. These factors, which implicate traditional districting principles, do correlate to some extent with the district’s layout. But we see no basis in the record for displacing the District Court’s conclusion that race predominated over them, particularly in light of the court’s findings that the State’s supporting data were not “available to the Legislature in any organized fashion before District 30 was created,” 861 F. Supp., at 1338, and that they do not “differentiate the district from surrounding areas,” ibid., with the same degree of correlation to district lines that racial data exhibit, see App. 150. In reaching that conclusion, we do not, as Justice Stevens fears, require States engaged in redistricting to compile “a comprehensive administrative record,” post, at 1026 (Stevens, J., dissenting), and we do not dismiss facts not explicitly mentioned in the redistricting plan’s legislative history as “irrelevant,” *967ibid. If, as may commonly happen, traditional districting principles are substantially followed without much conscious thought, they cannot be said to have been “subordinated to race.” In considering whether race was the “predominant factor motivating the legislature],” it is, however, eviden-tially significant that at the time of the redistricting, the State had compiled detailed racial data for use in redistricting, but made no apparent attempt to compile, and did not refer specifically to, equivalent data regarding communities of interest.
Appellants present a more substantial case for their claim that incumbency protection rivaled race in determining the district’s shape. Representative Johnson was the principal architect of District 30, which was designed in part to create a safe Democratic seat for her. At an early stage in the redistricting process, Johnson submitted to the state legislature a plan for Dallas County with a relatively compact 44% African-American district that did not violate the integrity of any voter tabulation district or county lines. See App. 139; 861 F. Supp., at 1338. The District Court found that “[w]hile minority voters did not object” to it, id., at 1330, “[t]hat plan drew much opposition from incumbents and was quickly abandoned,” id., at 1321, n. 22. “[F]ive other congressmen would have been thrown into districts other than the ones they currently represent.” Id., at 1330-1331. Appellants also point to testimony from Johnson and others to the effect that the incumbents of the adjacent Democratic Districts 5 and 24 exerted strong and partly successful efforts to retain predominantly African-American Democratic voters in their districts. (There was evidence that 97% of African-American voters in and around the city of Dallas vote Democrat.) See generally id., at 1321-1322.
In some circumstances, incumbency protection might explain as well as, or better than, race a State’s decision to depart from other traditional districting principles, such as compactness, in the drawing of bizarre district lines. And *968the fact that, “[a]s it happens, . . . many of the voters being fought over [by the neighboring Democratic incumbents] were African-American,” id., at 1338, would not, in and of itself, convert a political gerrymander into a racial gerrymander, no matter how conscious redistricters were of the correlation between race and party affiliation. See Shaw I, 509 U. S., at 646. If district lines merely correlate with race because they are drawn on the basis of political affiliation, which correlates with race, there is no racial classification to justify, just as racial disproportions in the level of prosecutions for a particular crime may be unobjectionable if they merely reflect racial disproportions in the commission of that crime, cf. post, at 1032, n. 30 (Stevens, J., dissenting) (discussing United States v. Armstrong, ante, at 456.
If the State’s goal is otherwise constitutional political gerrymandering, it is free to use the kind of political data on which Justice Stevens focuses — precinct general election voting patterns, post, at 1030, precinct primary voting patterns, post, at 1017, and legislators’ experience, post, at 1026 — to achieve that goal regardless of its awareness of its racial implications and regardless of the fact that it does so in the context of a majority-minority district. To the extent that the District Court suggested the contrary, it erred. But to the extent that race is used as a proxy for political characteristics, a racial stereotype requiring strict scrutiny is in operation. Cf. Powers v. Ohio, 499 U. S. 400, 410 (1991) (“Race cannot be a proxy for determining juror bias or competence”). We cannot agree with the dissenters, see post, at 1031 (Stevens, J., dissenting); post, at 1051-1052, n. 5 (Souter, J., dissenting); see also Shaw II, ante, at 924-925, n. 4 (Stevens, J., dissenting), that racial stereotyping that we have scrutinized closely in the context of jury service can pass without justification in the context of voting. If the promise of the Reconstruction Amendments, that our Nation is to be free of state-sponsored discrimination, is to be upheld, we cannot pick and choose between the basic forms of *969political participation in our efforts to eliminate unjustified racial stereotyping by government actors.
Here, the District Court had ample bases on which to conclude both that racially motivated gerrymandering had a qualitatively greater influence on the drawing of district lines than politically motivated gerrymandering, and that political gerrymandering was accomplished in large part by the use of race as a proxy. The State’s own VRA § 5 submission explains the drawing of District 30, and the rejection of Johnson’s more compact plan, in exclusively racial terms:
“Throughout the course of the Congressional redistricting process, the lines of the proposed District 30 were constantly reconfigured in an attempt to maximize the voting strength for this black community in Dallas County ... While the legislature was in agreement that a safe black district should be drawn in the Dallas County area, the real dispute involved the composition, configuration and quality of that district. The community insisted that [a] ‘safe’ black district be drawn that had a total black population of at least 50%.. ..
“. . . Although some [alternative] proposals showed a more compact configuration, none of them reached the threshold 50% total black population which the community felt was necessary to assure its ability to elect its own Congressional representative without having to form coalitions with other minority groups. .. .
“... The goal was to not only create a district that would maximize the opportunity for the black community to elect a Congressional candidate of its choice in 1992, but also one that included some of the major black growth areas which will assure continued electoral and economic opportunities over the next decades.” App. 106-107.
As the District Court noted, testimony of state officials in earlier litigation (in which District 30 was challenged as a political gerrymander) contradicted part of their testimony *970here, and affirmed that “race was the primary consideration in the construction of District 30.” 861 F. Supp., at 1338; see also id., at 1319-1321. And Johnson explained in a letter to the Department of Justice written at the end of the redistricting process that incumbency protection had been achieved by using race as a proxy:
“ ‘Throughout the course of the Congressional re district-ing process, the lines were continuously reconfigured to assist in protecting the Democratic incumbents in the Dallas/Fort Worth metroplex area by spreading the Black population to increase the Democratic party index in those areas.’” Id., at 1322 (quoting Plaintiff Exh. 6E6).
This is not to say that the direct evidence of the districters’ intent showed race to be the sole factor considered. As Justice Stevens notes, post, at 1024-1025, nn. 23-24, state officials’ claims have changed as their interests have changed. In the prior political gerrymandering suit and to the Department of Justice, they asserted that race predominated. In this suit, their testimony was that political considerations predominated. These inconsistent statements must be viewed in light 'of their adversarial context. But such questions of credibility are matters for the District Court, and we simply differ from the dissenters in our reading of the record when they find insupportable the District Court’s reliance on the State’s own statements indicating the importance of race, see post, at 1024-1025, nn. 23-24,1033, n. 31 (opinion of Stevens, J.).
Finally, and most significantly, the objective evidence provided by the district plans and demographic maps suggests strongly the predominance of race. Given that the district-ing software used by the State provided only racial data at the block-by-block level, the fact that District 30, unlike Johnson’s original proposal, splits voter tabulation districts and even individual streets in many places, see App. 150; 861 *971F. Supp., at 1339, suggests that racial criteria predominated over other districting criteria in determining the district’s boundaries. And, despite the strong correlation between race and political affiliation, the maps reveal that political considerations were subordinated to racial classification in the drawing of many of the most extreme and bizarre district lines. For example, the northernmost hook of the district, where it ventures into Collin County, is tailored perfectly to maximize minority population, see App. 153 (all whole and parts of 1992 voter tabulation districts within District 30’s Collin County hook have a combined African-American and Hispanic population in excess of 50%, with an average African-American population of 19.8%, id., at 331, while the combined African-American and Hispanic population in all surrounding voter tabulation districts, and the other parts of split districts, in Collin County is less than 25%), whereas it is far from the shape that would be necessary to maximize the Democratic vote in that area, see id., at 196 (showing a Republican majority, based on 1990 voting patterns in seven of the eight 1990 voter tabulation districts wholly or partly included in District 30 in Collin County).*
*972The combination of these factors compels us to agree with the District Court that “the contours of Congressional District 30 are unexplainable in terms other than race.” 861 F. Supp., at 1339. It is true that District 30 does not evince a consistent, single-minded effort to “segregate” voters on the basis of race, post, at 1023 (Stevens, J., dissenting), and does not represent “apartheid,” post, at 1054, 1074 (Souter, J., dissenting). But the fact that racial data were used in complex ways, and for multiple objectives, does not mean that race did not predominate over other considerations. The record discloses intensive and pervasive use of race both *973as a proxy to protect the political fortunes of adjacent incumbents, and for its own sake in maximizing the minority population of District 30 regardless of traditional districting principles. District 30’s combination of a bizarre, noncompact shape and overwhelming evidence that that shape was essentially dictated by racial considerations of one form or another is exceptional; Texas Congressional District 6, for example, which Justice Stevens discusses in detail, post, at 1019-1020, has only the former characteristic. That combination of characteristics leads us to conclude that District 30 is subject to strict scrutiny.
B
In Harris County, centered on the city of Houston, Districts 18 and 29 interlock “like a jigsaw puzzle ... in which it might be impossible to get the pieces apart.” Barone & Ujifusa, Almanac of American Politics 1996, at 1307-1308; see also Appendixes B and C to this opinion (outlines of Districts 18, 29). As the District Court noted: “[T]hese districts are so finely ‘crafted’ that one cannot visualize their exact boundaries without looking at a map at least three feet square.” 861 F. Supp., at 1323. According to the leading statistical study of relative district compactness and regularity, they are two of the three least regular districts in the country. See Pildes & Niemi, 92 Mich. L. Rev., at 565.
District 18’s population is 51% African-American and 15% Hispanic. App. 110. It “has some of the most irregular boundaries of any congressional district in the countryt,]... boundaries that squiggle north toward Intercontinental Airport and northwest out radial highways, then spurt south on one side toward the port and on the other toward the Astrodome.” Barone & Ujifusa, supra, at 1307. Its “many narrow corridors, wings, or fingers . . . reach out to enclose black voters, while excluding nearby Hispanic residents.” Pildes & Niemi, supra, at 556.
District 29 has a 61% Hispanic and 10% African-American population. App. 110. It resembles
*974“ ‘a sacred Mayan bird, with its body running eastward along the Ship Channel from downtown Houston until the tail terminates in Baytown. Spindly legs reach south to Hobby Airport, while the plumed head rises northward almost to Intercontinental. In the western extremity of the district, an open beak appears to be searching for worms in Spring Branch. Here and there, ruffled feathers jut out at odd angles.’” Barone & Ujifusa, supra, at 1335.
Not only are the shapes of the districts bizarre; they also exhibit utter disregard of city limits, local election precincts, and voter tabulation district lines. See, e. g., 861 F. Supp., at 1340 (60% of District 18 and District 29 residents live in split precincts). This caused a severe disruption of traditional forms of political activity. Campaigners seeking to visit their constituents “had to carry a map to identify the district lines, because so often the borders would move from block to block”; voters “did not know the candidates running for office” because they did not know which district they lived in. Ibid. In light of Texas’ requirement that voting be arranged by precinct, with each precinct representing a community that shares local, state, and federal representatives, it also created administrative headaches for local election officials:
“The effect of splitting dozens of [voter tabulation districts] to create Districts 18 and 29 was an electoral nightmare. Harris County estimated that it must increase its number of precincts from 672 to 1,225 to accommodate the new Congressional boundaries. Polling places, ballot forms, and the number of election employees are correspondingly multiplied. Voters were thrust into new and unfamiliar precinct alignments, a few with populations as low as 20 voters.” Id., at 1325.
See also App. 119-127 (letter from local official setting forth administrative problems and conflict with local districting *975traditions); id., at 147 (map showing splitting of city limits); id., at 128, Plaintiffs’ Exh. 6E1, Attachment A (map illustrating splitting of voting precincts).
As with District 30, appellants adduced evidence that incumbency protection played a role in determining the bizarre district lines. The District Court found that one constraint on the shape of District 29 was the rival ambitions of its two “functional incumbents,” who distorted its boundaries in an effort to include larger areas of their existing state legislative constituencies. 861 F. Supp., at 1340. But the District Court’s findings amply demonstrate that such influences were overwhelmed in the determination of the districts’ bizarre shapes by the State’s efforts to maximize racial divisions. The State’s VRA § 5 submission explains that the bizarre configuration of Districts 18 and 29 “result[s] in the maximization of minority voting strength” in Harris County, App. 110, corroborating the District Court’s finding that “[i]n the earliest stages of the Congressional redistricting process, state Democratic and Republican leaders rallied behind the idea of creating a new Hispanic safe seat in Harris County while preserving the safe African-American seat in District 18.” 861 F. Supp., at 1324. State officials testified that “it was particularly necessary to split [voter tabulation districts] in order to capture pockets of Hispanic residents” for District 29, and that a 61% Hispanic population in that district— not a mere majority — was insisted upon. Id., at 1340-1341. The record evidence of the racial demographics and voting patterns of Harris County residents belies any suggestion that party politics could explain the dividing lines between the two districts: The district lines correlate almost perfectly with race, see App. 151-152, while both districts are similarly solidly Democratic, see id., at 194. And, even more than in District 30, the intricacy of the lines drawn, separating Hispanic voters from African-American voters on a block-by-block basis, betrays the critical impact of the block-by-block racial data available on the REDAPPL program. The Dis*976trict Court’s conclusion is, therefore, inescapable: “Because Districts 18 and 29 are formed in utter disregard for traditional redistricting criteria and because their shapes are ultimately unexplainable on grounds other than the racial quotas established for those districts, they are the product of [presumptively] unconstitutional racial gerrymandering.” 861 F. Supp., at 1341.
Ill
Having concluded that strict scrutiny applies, we must determine whether the racial classifications embodied in any of the three districts are narrowly tailored to further a compelling state interest. Appellants point to three compelling interests: the interest in avoiding liability under the “results” test of VRA § 2(b), the interest in remedying past and present racial discrimination, and the “nonretrogression” principle of VRA § 5 (for District 18 only). We consider them in turn.
A
Section 2(a) of the VRA prohibits the imposition of any electoral practice or procedure that “results in a denial or abridgement of the right of any citizen ... to vote on account of race or color.” In 1982, Congress amended the VRA by changing the language of § 2(a) and adding § 2(b), which provides a “results” test for violation of §2(a). A violation exists if,
“based on the totality of circumstances, it is shown that the political processes leading to nomination or election in the State or political subdivision are not equally open to participation by members of a class of citizens protected by subsection (a) of this section in that its members have less opportunity than other members of the electorate to participate in the political process and to elect representatives of their choice.” 42 U. S. C. § 1973(b).
*977Appellants contend that creation of each of the three majority-minority districts at issue was justified by Texas’ compelling state interest in complying with this results test.
As we have done in each of our previous cases in which this argument has been raised as a defense to charges of racial gerrymandering, we assume without deciding that compliance with the results test, as interpreted by our precedents, see, e. g., Growe v. Emison, 507 U. S. 25, 37-42 (1993), can be a compelling state interest. See Shaw II, ante, at 915; Miller, 515 U. S., at 920-921. We also reaffirm that the “narrow tailoring” requirement of strict scrutiny allows the States a limited degree of leeway in furthering such interests. If the State has a “strong basis in evidence,” Shaw I, 509 U. S., at 656 (internal quotation marks omitted), for concluding that creation of a majority-minority district is reasonably necessary to comply with § 2, and the districting that is based on race “substantially addresses the § 2 violation,” Shaw II, ante, at 918, it satisfies strict scrutiny. We thus reject, as impossibly stringent, the District Court’s view of the narrow tailoring requirement, that “a district must have the least possible amount of irregularity in shape, making allowances for traditional districting criteria.” 861 F. Supp., at 1343. Cf. Wygant v. Jackson Bd. of Ed., 476 U. S. 267, 291 (1986) (O’Connor, J., concurring in part and concurring in judgment) (state actors should not be “trapped between the competing hazards of liability” by the imposition of unattainable requirements under the rubric of strict scrutiny).
A §2 district that is reasonably compact and regular, taking into account traditional districting principles such as maintaining communities of interest and traditional boundaries, may pass strict scrutiny without having to defeat rival compact districts designed by plaintiffs’ experts in endless “beauty contests.” The dissenters misread us when they make the leap from our disagreement about the facts of this *978suit to the conclusion that we are creating a “stalemate” by requiring the States to “get things just right,” post, at 1063 (Souter, J., dissenting), or to draw “the precise compact district that a court would impose in a successful §2 challenge,” post, at 1035 (Stevens, J., dissenting); see also Shaw II, ante, at 949 (Stevens, J., dissenting). Rather, we adhere to our longstanding recognition of the importance in our federal system of each State’s sovereign interest in implementing its redistricting plan. See Voinovich v. QuiLter, 507 U. S. 146, 156 (1993) (“[I]t is the domain of the States, and not the federal courts, to conduct apportionment in the first place”); Miller, supra, at 915 (“It is well settled that reapportionment is primarily the duty and responsibility of the State”) (internal quotation marks omitted). Under our cases, the States retain a flexibility that federal courts enforcing §2 lack, both insofar as they may avoid strict scrutiny altogether by respecting their own traditional districting principles, and insofar as deference is due to their reasonable fears of, and to their reasonable efforts to avoid, § 2 liability. And nothing that we say today should be read as limiting “a State’s discretion to apply traditional districting principles,” post, at 1046 (Souter, J., dissenting), in majority-minority, as in other, districts. The constitutional problem arises only from the subordination of those principles to race.
Strict scrutiny remains, nonetheless, strict. The State must have a “strong basis in evidence” for finding that the threshold conditions for § 2 liability are present:
“first, ‘that [the minority group] is sufficiently large and geographically compact to constitute a majority in a single member district’; second, ‘that it is politically cohesive’; and third, ‘that the white majority votes sufficiently as a bloc to enable it . . . usually to defeat the minority’s preferred candidate.’” Growe, supra, at 40 (emphasis added) (quoting Thornburg v. Gingles, 478 U. S. 30, 50-51 (1986)).
*979And, as we have noted above, the district dráwn in order to satisfy § 2 must not subordinate traditional districting principles to race substantially more than is “reasonably necessary” to avoid §2 liability. Districts 18, 29, and 30 fail to meet these requirements.
We assume, without deciding, that the State had a “strong basis in evidence” for finding the second and third threshold conditions for § 2 liability to be present. We have, however, already found that all three districts are bizarrely shaped and far. from compact, and that those characteristics are predominantly attributable to gerrymandering that was racially motivated and/or achieved by the use of race as a proxy. See Part II, supra. District 30, for example, reaches out to grab small and apparently isolated minority communities which, based on the evidence presented, could not possibly form part of a compact majority-minority district, and does so in order to make up for minority populations closer to its core that it shed in a further suspect use of race as a proxy to. further neighboring incumbents’ interests. See supra, at 965-966, 969-973.
These characteristics defeat any claim that the districts are narrowly tailored to serve the State’s interest in avoiding liability under §2, because §2 does not require a State to create, on predominantly racial lines, a district that is not “reasonably compact.” See Johnson v. De Grandy, 512 U. S. 997, 1008 (1994). If, because of the dispersion of the minority population, a reasonably compact majority-minority district cannot be created, §2 does not require a majority-minority district; if a reasonably compact district can be created, nothing in § 2 requires the race-based creation of a district that is far from compact.
Appellants argue that bizarre shaping and noncom-pactness do not raise narrow tailoring concerns. Appellants Lawson et al. claim that under Shaw I and Miller, “[s]hape is relevant only as evidence of an improper motive.” *980Brief for Appellants Lawson et al. 56. They rely on our statement in Miller:
“Shape is relevant not because bizarreness is a necessary element of the constitutional wrong or a threshold requirement of proof, but because it may be persuasive circumstantial evidence that race for its own sake, and not other districting principles, was the legislature’s dominant and controlling rationale in drawing its district lines.” 515 U. S., at 913.
The United States takes a more moderate position, accepting that in the context of narrow tailoring, “consideration must be given to the extent to which the districts drawn by a State substantially depart from its customary redistricting practices,” Brief for United States 36, but asserting that insofar as bizarreness and noncompactness are necessary to achieve the State’s compelling interest in compliance with § 2 “while simultaneously achieving other legitimate redistricting goals,” id., at 37, such as incumbency protection, the narrowly tailoring requirement is satisfied. Similarly, Justice Stevens’ dissent argues that “noncompact districts should . . . be a permissible method of avoiding violations of [§2J.” Post, at 1034.
These arguments cannot save the districts before us. The Lawson appellants misinterpret Miller: District shape is not irrelevant to the narrow tailoring inquiry. Our discussion in Miller served only to emphasize that the ultimate constitutional values at stake involve the harms caused by the use of unjustified racial classifications, and that bizarreness is not necessary to trigger strict scrutiny. See Miller, 515 U. S., at 912-913. Significant deviations from traditional district-ing principles, such as the bizarre shape and noncompactness demonstrated by the districts here, cause constitutional harm insofar as they convey the message that political identity is, or should be, predominantly racial. For example, the bizarre shaping of Districts 18 and 29, cutting across pre*981existing precinct lines and other natural or traditional divisions, is not merely evidentially significant; it is part of the constitutional problem insofar as it disrupts nonracial bases of political identity and thus intensifies the emphasis on race.
Nor is the United States’ argument availing here. In determining that strict scrutiny applies here, we agreed with the District Court that in fact the bizarre shaping and non-compactness of these districts were predominantly attributable to racial, not political, manipulation. The United States’ argument, and that of the dissent, post, at 1033-1035 (Stevens, J., dissenting), address the case of an otherwise compact majority-minority district that is misshapen by predominantly nonracial, political manipulation. See also post, at 1068 (Souter, J., dissenting) (raising “the possibility that a State could create a majority-minority district that does not coincide with the Gingles shape so long as racial data are not overused”). We disagree with the factual premise of Justice Stevens’ dissent, that these districts were drawn using “racial considerations only in a way reasonably designed” to avoid a § 2 violation, post, at 1035. The districts before us exhibit a level of racial manipulation that exceeds what § 2 could justify.
B
The United States and the State next contend that the district lines at issue are justified by the State’s compelling interest in “ameliorating the effects of racially polarized voting attributable to past and present racial discrimination.” Brief for United States 32; Brief for Appellants Bush et al. 24-25. In support of that contention, they cite Texas’ long history of discrimination against minorities in electoral processes, stretching from the Reconstruction to modern times, including violations of the Constitution and of the VRA. See, e. g., Williams v. Dallas, 734 F. Supp. 1317 (ND Tex. 1990); White v. Regester, 412 U. S. 755 (1973); Terry v. Adams, 345 U. S. 461 (1953); Smith v. Allwright, 321 U. S. 649 (1944); Nixon v. Condon, 286 U. S. 73 (1932); Nixon v. *982Herndon, 273 U. S. 536 (1927); see also 861 F. Supp., at 1317 (because of its history of official discrimination, Texas became a covered jurisdiction under VRA § 5 in 1975, and the Department of Justice has since “frequently interposed objections against the State and its subdivisions”). Appellants attempt to link that history to evidence that in recent elections in majority-minority districts, “Anglos usually bloc voted against” Hispanic and African-American candidates. Ibid.
A State’s interest in remedying discrimination is compelling when two conditions are satisfied. First, the discrimination that the State seeks to remedy must be specific, “identified discrimination”; second, the State “must have had a ‘strong basis in evidence’ to conclude that remedial action was necessary, ‘before it embarks on an affirmative action program.’” Shaw II, ante, at 910 (citations omitted). Here, the only current problem that appellants cite as in need of remediation is alleged vote dilution as a consequence of racial bloc voting, the same concern that underlies their VRA §2 compliance defense, which we have assumed to be valid for purposes of this opinion. We have indicated that such problems will not justify race-based districting unless “the State employ[s] sound districting principles, and ... the affected racial group’s residential patterns afford the opportunity of creating districts in which they will be in the majority.” Shaw I, 509 U. S., at 657 (internal quotation marks omitted). Once that standard is applied, our agreement with the District Court’s finding that these districts are not narrowly tailored to comply with §2 forecloses this line of defense.
C
The final contention offered by the State and private appellants is that creation of District 18 (only) was justified by a compelling state interest in complying with VRA § 5. We have made clear that § 5 has a limited substantive goal: “ ‘to insure that no voting-procedure changes would be made that *983would lead to a retrogression in the position of racial minorities with respect to their effective exercise of the electoral franchise.’ ” Miller, 515 U. S., at 926 (quoting Beer v. United States, 425 U. S. 130, 141 (1976)). Appellants contend that this “nonretrogression” principle is implicated because Harris County had, for two decades, contained a congressional district in which African-American voters had succeeded in selecting representatives of their choice, all of whom were African-Americans.
The problem with the State’s argument is that it seeks to justify not maintenance, but substantial augmentation, of the African-American population percentage in District 18. At the previous redistricting, in 1980, District 18’s population was 40.8% African-American. Plaintiffs’ Exh. 13B, p. 55. As a result of Hispanic population increases and African-American emigration from the district, its population had reached 35.1% African-American and 42.2% Hispanic at the time of the 1990 census. The State has shown no basis for concluding that the increase to a 50.9% African-American population in 1991 was necessary to ensure nonretrogression. Nonretrogression is not a license for the State to do whatever it deems necessary to ensure continued electoral success; it merely mandates that the minority’s opportunity to elect representatives of its choice not be diminished, directly or indirectly, by the State’s actions. We anticipated this problem in Shaw I, 509 U. S., at 655: “A reapportionment plan would not be narrowly tailored to the goal of avoiding retrogression if the State went beyond what was reasonably necessary to avoid retrogression.” Applying that principle, it is clear that District 18 is not narrowly tailored to the avoidance of § 5 liability.
IV
The dissents make several further arguments against today’s decision, none of which address the specifics of this case. We have responded to these points previously. Justice Souter, for example, reiterates his contention from *984Shaw I that because districts created with a view to satisfying §2 do not involve “racial subjugation,” post, at 1055, and may in a sense be “‘benignly]’” motivated, Shaw I, 509 U. S., at 685 (Souter, J., dissenting), strict scrutiny should not apply to them. We rejected that argument in Shaw I, and we reject it now. As we explained then, see id., at 653, we subject racial classifications to strict scrutiny precisely because that scrutiny is necessary to determine whether they are benign — as Justice Stevens’ hypothetical of a targeted outreach program to protect victims of sickle cell anemia, see post, at 1032, would, no doubt, be — or whether they misuse race and foster harmful and divisive stereotypes without a compelling justification. We see no need to revisit our prior debates.
Both dissents contend that the recognition of the Shaw I cause of action threatens public respect for, and the independence of, the Federal Judiciary by inserting the courts deep into the districting process. We believe that the dissents both exaggerate the dangers involved, and fail to recognize the implications of their suggested retreat from Shaw I.
As to the dangers of judicial entanglement, Justice Stevens’ dissent makes much of cases stemming from state dis-tricting plans originally drawn up before Shaw I, in which problems have arisen from the uncertainty in the law prior to and during its gradual clarification in Shaw I, Miller, and today’s cases. See post, at 1037-1038 (Stevens, J., dissenting). We are aware of the difficulties faced by the States, and by the district courts, in confronting new constitutional precedents, and we also know that the nature of the expressive harms with which we are dealing, and the complexity of the districting process, are such that bright-line rules are not available. But we believe that today’s decisions, which both illustrate the defects that offend the principles of Shaw I and reemphasize the importance of the States’ discretion in the redistricting process, see supra, at 978-979, will serve *985to clarify the States’ responsibilities. The States have traditionally guarded their sovereign districting prerogatives jealously, and we are confident that they can fulfill that requirement, leaving the courts to their customary and appropriate backstop role.
This Court has now rendered decisions after plenary consideration in five cases applying the Shaw I doctrine (Shaw I, Miller, Hays, Shaw II, and this suit). The dissenters would have us abandon those precedents, suggesting that fundamental concerns relating to the judicial role are at stake. See post, at 1035, 1038, 1041 (Stevens, J., dissenting); post, at 1047, and n. 2,1052,1064,1074,1076-1077 (Sou-TER, J., dissenting); Shaw II, ante, at 919-920, 922-923, and n. 3, 929 (Stevens, J., dissenting); but see ante, at 932-933 (noting that the judicial task of distinguishing race-based from non-race-based action in Shaw I cases is far from unique). While we agree that those concerns are implicated here, we believe they point the other way. Our legitimacy requires, above all, that we adhere to stare decisis, especially in such sensitive political contexts as the present, where partisan controversy abounds. Legislators and district courts nationwide have modified their practices — or, rather, reem-braced the traditional districting practices that were almost universally followed before the 1990 census — in response to Shaw I. Those practices and our precedents, which acknowledge voters as more than mere racial statistics, play an important role in defining the political identity of the American voter. Our Fourteenth Amendment jurisprudence evinces a commitment to eliminate unnecessary and excessive governmental use and reinforcement of racial stereotypes. See, e. g., Georgia v. McCollum, 505 U. S. 42, 59 (1992) (“[T]he exercise of a peremptory challenge must not be based on either the race of the juror or the racial stereotypes held by the party”); Edmonson v. Leesville Concrete Co., 500 U. S. 614, 630-631 (1991) (“If our society is to continue to progress as a multiracial democracy, it must recog*986nize that the automatic invocation of race stereotypes retards that progress and causes continued hurt and injury”); Powers, 499 U. S., at 410 (“We may not accept as a defense to racial discrimination the very stereotype the law condemns”); Holland v. Illinois, 493 U. S. 474, 484, n. 2 (1990) (“[A] prosecutor’s 'assumption that a black juror may be presumed to be partial simply because he is black’ . . . violates the Equal Protection Clause”); Batson v. Kentucky, 476 U. S. 79, 104 (1986) (“[T]he Equal Protection Clause prohibits a State from taking any action based on crude, inaccurate racial stereotypes”). We decline to retreat from that commitment today.
* * *
The judgment of the District Court is
Affirmed.
*987APPENDIX A TO OPINION OF O’CONNOR, J.
[[Image here]]
TEXAS CONGRESSIONAL DISTRICT 30
*988APPENDIX B TO OPINION OF O’CONNOR, J.
[[Image here]]
TEXAS CONGRESSIONAL DISTRICT 18
*989APPENDIX C TO OPINION OF O’CONNOR, J.
[[Image here]]
TEXAS CONGRESSIONAL DISTRICT 29
In the application of our precedents to District 30, our disagreement with Justice Stevens’ dissent, post, at 1014-1031, is largely factual. In reviewing the District Court’s findings of primary fact, we cannot ignore the reality that the District Court heard several days of testimony and argument and became significantly more familiar with the factual details of this suit than this Court can be. We therefore believe that the dissent errs in second-guessing the District Court’s assessment of the witnesses’ testimony, see post, at 1025, n. 24, and in dismissing as mere “fine tuning,” post, at 1030, the practice of using race as a proxy that the District Court found, based on ample evidence, to be pervasive, see Vera v. Richards, 861 F. Supp 1304, 1322 (SD Tex. 1994).
For the same reason, we decline to debate the dissent on every factual nuance on which it diverges from the District Court’s, and our, .view. But two of its specific claims about District 30 merit a response. First, the dissent asserts that “[a] comparison of the 1992 precinct results with a depiction of the proportion of black population in each census block reveals that Democratic-leaning precincts cover a far greater area [of District 30] *972than majority-black census blocks.” Post, at 1030 (emphasis added). While that may be true, the dissent’s reliance on 1992 election results is misplaced. Those results were not before the legislature when it drew the district lines in 1991, and may well reflect the popularity and campaign success of Representative Johnson more than the party political predispositions of the district’s residents. (The same error infects the dissent’s discussion of the Collin County hook, post, at 1020-1021, n. 19 (relying on 1992 election results).) And looking at totals, rather than at the difference between areas just inside and just outside the district lines, is misleading: Race may predominate in the drawing of district lines because those lines are finely drawn to maximize the minority composition of the district, notwithstanding that in an overwhelmingly Democratic area, the total of Democrats in the district far exceeds its total minority population.
Second, the dissent suggests that strict scrutiny should not apply because District 30’s compact core has a higher African-American population percentage than its wayward tentacles. Post, at 1021-1023. In doing so, it again ignores the necessity of determining whether race predominated in the redistricters’. actions in light of what they had to work with. Once various adjacent majority-minority populations had been carved away from it by the use of race as a proxy to enhance the electoral chances of neighboring incumbents, the core of District 30 was substantially too small to form an entire district. The principal question faced by the redistrict-ers was, therefore, what territory to add to the core out of the remainder of the Dallas area, which remainder has an average African-American population substantially below the 21% county average. In answering that question, as the District Court explained and the maps bear witness, the redistricters created bizarre, far-reaching tentacles that intricately and consistently maximize the available remaining African-American population.