dissenting.
This case concerns the congressional reapportionment of New Jersey. The districting plan enacted by the New Jersey Legislature and signed into law by the Governor on January 19, 1982, Pub. L. 1982, ch. 1, reduced the number of congressional districts in the State from 15 to 14 as required by the 1980 census figures. The 14 congressional districts created by the legislature have an average deviation of 0.1384% and a maximum deviation between the largest and smallest districts of 0.6984%. In other words, this case concerns a *766maximum difference of 3,674 individuals in districts encompassing more than a half million people. The New Jersey plan was invalidated by a divided District Court because these population variances were not “ ‘unavoidable despite a good-faith effort to achieve absolute equality.’” Daggett v. Kimmelman, 535 F. Supp. 978, 982 (NJ 1982), quoting Kirkpatrick v. Preisler, 394 U. S. 526, 531 (1969). Today, the Court affirms the District Court’s decision thereby striking for the first time in the Court’s experience a legislative or congressional districting plan with an average and maximum population variance of under 1%.
I respectfully dissent from the Court’s unreasonable insistence on an unattainable perfection in the equalizing of congressional districts. The Court’s decision today is not compelled by Kirkpatrick v. Preisler, supra, and White v. Weiser, 412 U. S. 783 (1973), see Part I, infra, and if the Court is convinced that our cases demand the result reached today, the time has arrived to reconsider these precedents. In any event, an affirmance of the decision below is inconsistent with the majority’s own “modifications” of Kirkpatrick and White which require, at a minimum, further consideration of this case by the District Court. See Part IV, infra.
“[T]he achieving of fair and effective representation for all citizens is concededly the basic aim of legislative apportionment.” Reynolds v. Sims, 377 U. S. 533, 565-566 (1964). One must suspend credulity to believe that the Court’s draconian response to a trifling 0.6984% maximum deviation promotes “fair and effective representation” for the people of New Jersey. The requirement that “as nearly as is practicable one man’s vote in a congressional election is to be worth as much as another’s,” Wesberry v. Sanders, 376 U. S. 1, 7-8 (1964), must be understood in light of the malapportionment in the States at the time Wesberry was decided. The plaintiffs in Wesberry were voters in a congressional district (population 823,680) encompassing Atlanta that was three *767times larger than Georgia’s smallest district (272,154) and more than double the size of an average district. Because the State had not reapportioned for 30 years, the Atlanta District possessing one-fifth of Georgia’s population had only one-tenth of the Congressmen. Georgia was not atypical; congressional districts throughout the country had not been redrawn for decades and deviations of over 50% were the rule.1 These substantial differences in district size diminished, in a real sense, the representativeness of congressional elections. The Court’s invalidation of these profoundly unequal districts should not be read as a demand for precise mathematical equality between the districts. Indeed, the Court sensibly observed that “it may not be possible [for the States] to draw congressional districts with mathematical precision.” Id., at 18. In Reynolds v. Sims, supra, at 577, decided the same Term, the Court disavowed a requirement of mathematical exactness for legislative districts in even more explicit terms:
“We realize that it is a practical impossibility to arrange legislative districts so that each one has an identical number of residents, or citizens, or voters. Mathematical exactness or precision is hardly a workable constitutional requirement.”
The States responded to Wesberry by eliminating gross disparities between congressional districts. Nevertheless, redistricting plans with far smaller variations were struck by the Court five years later in Kirkpatrick v. Preisler, supra, and its companion, Wells v. Rockefeller, 394 U. S. 542 (1969). The redistricting statutes before the Court contained total percentage deviations of 5.97% and 13.1%, respectively. *768But Wesberry’s “as nearly as practicable” standard was read to require “a good-faith effort to achieve precise numerical equality.” 394 U. S., at 530-531. Over the objections of four Justices, see id., at 536 (Fortas, J., concurring); id., at 549 (Harlan, J., joined by Stewart, J., dissenting); id., at 553 (White, J., dissenting), Kirkpatrick rejected the argument that there is a fixed numerical or percentage population variance small enough to be considered de minimis and to satisfy the “as nearly as practicable” standard. Kirkpatrick’s, rule was applied by the Court in White v. Weiser, supra, to invalidate Texas’ redistricting scheme which had a maximum population variance of 4.13%.
Just as Wesberry did not require Kirkpatrick, Kirkpatrick does not ineluctably lead to the Court’s decision today. Although the Court stated that it could see “no nonarbitrary way” to pick a de minimis point, the maximum deviation in Kirkpatrick, while small, was more than eight times as large as that posed here. Moreover, the deviation in Kirkpatrick was not argued to fall within the officially accepted range of statistical imprecision of the census. Interestingly enough, the Missouri redistricting plan approved after Kirkpatrick contained a deviation of 0.629% — virtually the same deviation declared unconstitutional in this case. Preisler v. Secretary of State of Missouri, 341 F. Supp. 1158, 1162 (WD Mo.), summarily aff’d sub nom. Danforth v. Preisler, 407 U. S. 901 (1972).2 Accordingly, I do not view the Court’s decision today as foreordained by Kirkpatrick and Weiser. Apparently neither did Justice Brennan who, in staying the District Court’s order, wrote:
“The appeal would thus appear to present the important question whether Kirkpatrick v. Preisler requires adoption of the plan that achieves the most precise math*769ematical exactitude, or whether Kirkpatrick left some latitude for the New Jersey Legislature to recognize the considerations taken into account by it as a basis for choosing among several plans, each with arguably ‘statistically insignificant’ variances from the constitutional ideal of absolute precision.” 455 U. S. 1303,1305 (1982).
There can be little question but that the variances in the New Jersey plan are “statistically insignificant.” Although the Government strives to make the decennial census as accurate as humanly possible, the Census Bureau has never intimated that the results are a perfect count of the American population. The Bureau itself estimates the inexactitude in the taking of the 1970 census at 2.3%,3 a figure which is considerably larger than the 0.6984% maximum variance in the New Jersey plan, and which dwarfs the 0.2470% difference between the maximum deviations of the selected plan and the leading alternative plan, that suggested by Professor Reock. Because the amount of undercounting differs from district to district, there is no point for a court of law to act under an unproved assumption that such tiny differences between redistricting plans reflect actual differences in population. As Dr. James Trussel, an expert in these matters, and whose testimony the Court purports to accept, ante, at 735-736, explained:
“The distribution of the undercount in New Jersey is obviously also unknown, and I see no reason to believe that *770it would be uniformly spread over all municipalities. For these reasons, one cannot make congressional districts of truly equal size if one relies on census counts. Nor is it meaningful to rank one redistricting plan as superior to another when differences in district size are small. In my professional opinion, districts whose enumerated populations differ one from another by less than one percent should,be considered to be equal in size. To push for numerical equality beyond this point is an exercise in illusion.” App. 103-104.4
*771Even if the 0.6984% deviation here is not encompassed within the scope of the statistical imprecision of the census, it is miniscule when compared with other variations among the districts inherent in translating census numbers into citizens’ votes. First, the census “is more of an event than a process.” Gaffney v. Cummings, 412 U. S. 735, 746 (1973). “It measures population at only a single instant in time. District populations are constantly changing, often at different rates in either direction, up or down.” Ibid. As the Court admits, “the well-known restlessness of the American people means that population counts for particular localities are outdated long before they are completed.” Ante, at 732.5 Second, far larger differences among districts are introduced because a substantial percentage of the total population is too *772young to register or is disqualified by alienage.6 Third, census figures cannot account for the proportion of all those otherwise eligible individuals who fail to register.7 The differences in the number of eligible voters per district for these reasons overwhelm the minimal variations attributable to the districting plan itself.8
Accepting that the census, and the districting plans which are based upon it, cannot be perfect represents no backsliding in our commitment to assuring fair and equal representation in the election of Congress. I agree with the views of Judge Gibbons, who dissented in the District Court, that Kirkpatrick should not be read as a “prohibition against toleration of de minimis population variances which have no statistically relevant effect on relative representation.” Daggett v. Kimmelman, 535 F. Supp., at 984. A plus-minus deviation of 0.6984% surely falls within this category.
If today’s decision simply produced an unjustified standard with little practical import, it would be bad enough. Unfortunately, I fear that the Court’s insistence that “there are no de minimis population variations, which could practicably be avoided, but which nonetheless meet the standard of Art. I, § 2, without justification,” ante, at 734, invites further litigation of virtually every congressional redistricting plan in *773the Nation. At least 12 States which have completed redistricting on the basis of the 1980 census have adopted plans with a higher deviation than that presented here, and 4 others have deviations quite similar to New Jersey’s.9 Of course, under the Court’s rationale, even Rhode Island’s plan — whose two districts have a deviation of 0.02% or about 95 people — would be subject to constitutional attack.
In all such cases, state legislatures will be hard pressed to justify their preference for the selected plan. A good-faith effort to achieve population equality is not enough if the population variances are not “unavoidable.” The court must consider whether the population differences could have been further “reduced or eliminated altogether.” Ante, at 730. With the assistance of computers, there will generally be a plan with an even more minimal deviation from the mathematical ideal. Then, “the State must bear the burden of proving that each significant variance between districts was necessary to achieve some legitimate goal.” Ante, at 731. As this case illustrates, literally any variance between districts will be considered “significant.”10 The State’s burden will not be easily met: “the State bears the burden of justifying *774the differences with particularity.” Ante, at 7B9. When the State fails to sustain its burden, the result will generally be that a court must select an alternative plan. The choice will often be disputed until the very eve of an election, see, e. g., Upham v. Seamon, 456 U. S. 37, 44 (1982) (per curiam), leaving candidates and voters in a state of confiision.
The only way a legislature or bipartisan commission can hope to avoid litigation will be to dismiss all other legitimate concerns and opt automatically for the districting plan with the smallest deviation.11 Yet no one can seriously contend that such an inflexible insistence upon mathematical exactness will serve to promote “fair and effective representation.” The more likely result of today’s extension of Kirkpatrick is to move closer to fulfilling Justice Fortas’ prophecy that “a legislature might have to ignore the boundaries of common sense, running the congressional district line down the middle of the corridor of an apartment house or even dividing the residents of a single-family house between two districts.” 394 U. S., at 538. Such sterile and mechanistic application only brings the principle of “one man, one vote” into disrepute.
II
One might expect the Court had strong reasons to force this Sisyphean task upon the States. Yet the Court offers *775no positive virtues that will follow from its decision. No pretense is made that this case follows in the path of Reynolds and Wesberry in insuring the “fair and effective representation” of citizens. No effort is expended to show that Art. I, § 2’s requirement that Congressmen be elected “by the people,” Wesberry v. Sanders, 376 U. S. 1 (1964), demands the invalidation of population deviations at this level. Any such absolute requirement, if it did exist, would be irreconcilable with the Court’s recognition of certain justifications for population variances. See ante, at 740. Given no express constitutional basis for the Court’s holding, and no showing that the objectives of fair representation are compromised by these minimal disparities, the normal course would be to uphold the actions of the legislature in fulfilling its constitutionally delegated responsibility to prescribe the manner of holding elections for Senators and Representatives. Art. I, § 4. Doing so would be in keeping with the Court’s oft-expressed recognition that apportionment is primarily a matter for legislative judgment. Upham v. Seamon, supra, at 41; White v. Weiser, 412 U. S., at 795; Reynolds v. Sims, 377 U. S., at 586. “[A] state legislature is the institution that is by far the best situated to identify and then reconcile traditional state policies within the constitutionally mandated framework .. ..” Connor v. Finch, 431 U. S. 407, 414-415 (1977).
Instead the.Court is purely defensive in support of its decision. The Court refiises to adopt any fixed numerical standard, below which the federal courts would not intervene, asserting that “[t]he principle of population equality for congressional districts has not proved unjust or socially or economically harmful in experience.” Ante, at 733. Of course, the principle of population equality is not unjust; the unreasonable application of this principle is the rub. Leaving aside that the principle has never been applied with the vengeance witnessed today, there are many, including myself, who take issue with the Court’s self-congratulatory assumption that Kirkpatrick has been a success. First, a *776decade of experience with Kirkpatrick has shown that “the rule of absolute equality is perfectly compatible with ‘gerrymandering’ of the worst sort.” Wells v. Rockefeller, 394 U. S., at 551 (Harlan, J., dissenting). With ever more sophisticated computers, legislators can draw countless plans for absolute population equality, but each having its own political ramifications. Although neither a rule of absolute equality nor one of substantial equality can alone prevent deliberate partisan gerrymandering, the former offers legislators a ready justification for disregarding geographical and political boundaries. I remain convinced of what I said in dissent in Kirkpatrick and Wells: “[Those] decisions . . . downgrade a restraint on a far greater potential threat to equality of representation, the gerrymander. Legislatures intent on minimizing the representation of selected political or racial groups are invited to ignore political boundaries and compact districts so long as they adhere to population equality among districts using standards which we know and they know are sometimes quite incorrect.” 349 U. S., at 555. There is now evidence that Justice Harlan was correct to predict that “[e]ven more than in the past, district lines are likely to be drawn to maximize the political advantage of the party temporarily dominant in public affairs.” Id., at 552.12
*777In addition to providing a patina of respectability for the equipopulous gerrymander, Kirkpatrick’s regime assured extensive intrusion of the judiciary into legislative business. *778“[T]he [re]apportionment task, dealing as it must with fundamental ‘choices about the nature of representation,' Burns v. Richardson, 384 U. S., at 92, is primarily a political and legislative process.” Gaffney v. Cummings, 412 U. S., at 749. What we said in Gaffney with respect to legislative reapportionment is apropos here:
“[T]he goal of fair and effective representation [is not] furthered by making the standards of reapportionment so difficult to satisfy that the reapportionment task is re-curringly removed from legislative hands and performed by federal courts which themselves must make the political decisions necessary to formulate a plan or accept those made by reapportionment plaintiffs who may have wholly different goals from those embodied in the official plan.” Ibid.
More than a decade’s experience with Kirkpatrick demonstrates that insistence on precise numerical equality only invites those who lost in the political arena to refight their battles in federal court. Consequently, “[m]ost estimates are that between 25 percent and 35 percent of current house district lines were drawn by the Courts.” American Bar Association, Congressional Redistricting 20 (1981). As I have already noted, by extending Kirkpatrick to deviations below even the 1% level, the redistricting plan in every State with more than a single Representative is rendered vulnerable to after-the-fact attack by anyone with a complaint and a calculator.
The Court ultimately seeks refuge in stare decisis. I do not slight the respect that doctrine is due, see, e. g., White v. *779Weiser, 412 U. S. 783 (1973), but is it not at least ironic to find stare decisis invoked to protect Kirkpatrick as the Court itself proceeds to overrule other holdings in that very decision? In Kirkpatrick, the Court squarely rejected the argument that slight variances in district size were proper in order to avoid fragmenting political subdivisions:
“[W]e do not find legally acceptable the argument that variances are justified if they necessarily result from a State’s attempt to avoid fragmenting political subdivisions by drawing congressional district lines along existing county, municipal, or other political subdivision boundaries.” 394 U. S., at 533-534.13
Several pages later, the Court rejected in equally uncategorical terms the idea that variances may be justified in order to make districts more compact. Id., at 535-536. “A State’s preference for pleasingly shaped districts,” the Court concluded, “can hardly justify population variances.” Id., at 536. In Justice Fortas’ words, the Kirkpatrick Court “reject^], seriatim, every type of justification that has been— possibly, every one that could be — advanced.” Id., at 537.
Yet today the Court — with no mention of the contrary holdings in Kirkpatrick — opines: “Any number of consistently applied legislative policies might justify some variance, including for instance, making districts compact, respecting municipal boundaries, preserving the cores of prior districts, and avoiding contests between incumbent Representatives.” *780Ante, at 740. I, of course, welcome the Court’s overruling of these ill-considered holdings of Kirkpatrick. There should be no question but that state legislatures may account for political and geographic boundaries in order to preserve traditional subdivisions and achieve compact and contiguous districts. Justice Stevens recognizes that courts should “give greater weight to the importance of the State’s interests and the consistency with which those interests are served than to the size of the deviations.” Ante, at 760, n. 26. Thus, a majority of the Court appears ready to apply this new standard “with a strong measure of deference to the legitimate concerns of the State.” Post, at 785, n. 1 (Powell, J., dissenting).
In order that legislatures have room to accommodate these legitimate noncensus factors, a range of de minimis population deviation, like that permitted in the legislative reapportionment cases, is required. The Court’s insistence that every deviation, no matter how small, be justified with specificity discourages legislatures from considering these “legitimate” factors in making their plans, lest the justification be found wanting, the plan invalidated, and a judicially drawn substitute put in its place. Moreover, the requirement of precise mathematical equality continues to invite those who would bury their political opposition to employ equipopulous gerrymanders. A de minimis range would not preclude such gerrymanders but would at least force the political cartographer to justify his work on its own terms.
r-H h-H
Our cases dealing with state legislative apportionment have taken a more sensible approach. We have recognized that certain small deviations do not, in themselves, ordinarily constitute a prima facie constitutional violation. Gaffney v. Cummings, 412 U. S. 735 (1973); White v. Regester, 412 U. S. 755 (1973). Moreover, we have upheld plans with reasonable variances that were necessary to account for political *781subdivisions, Mahan v. Howell, 410 U. S. 315 (1973), to preserve the voting strength of minority groups, and to insure political fairness, Gaffney v. Cummings, supra. What we held in Gaffney v. Cummings for legislative apportionment is fully applicable to congressional redistricting:
“ ‘[T]he achieving of fair and effective representation for all citizens is’... a vital and worthy goal, but surely its attainment does not in any commonsense way depend upon eliminating the insignificant population variations involved in this case. Fair and effective representation may be destroyed by gross population variations among districts, but it is apparent that such representation does not depend solely on mathematical equality among district populations .... An unrealistic overemphasis on raw population figures, a mere nose count in the districts, may submerge these other considerations and itself furnish a ready tool for ignoring factors that in day-to-day operation are important to an acceptable representation and apportionment arrangement.” 412 U.S., at 748-749.
Bringing together our state legislative and congressional cases does not imply overlooking relevant differences between the two. States normally draw a larger number of legislative districts, which accordingly require a greater margin to account for geographical and political boundaries. “[Congressional districts are not so intertwined and freighted with strictly local interests as are state legislative districts.” White v. Weiser, 412 U. S., at 793. Furthermore, because congressional districts are generally much larger than state legislative districts, each percentage point of variation represents a commensurately greater number of people. But these are differences of degree. They suggest that the level at which courts should entertain challenges to districting plans, absent unusual circumstances, should be lower in the *782congressional cases, but not altogether nonexistent.14 Although I am not wedded to a precise figure, in light of the current range of population deviations, a 5% cutoff appears reasonable. I would not entertain judicial challenges, absent extraordinary circumstances, where the maximum deviation is less than 5%. Somewhat greater deviations, if rationally related to an important state interest, may also be permissible.15 Certainly, the maintaining of compact, contiguous districts, the respécting of political subdivisions, and efforts to assure political fairness, e. g., Gaffney v. Cummings, supra, constitute such interests.
I would not hold up New Jersey’s plan as a model reflection of such interests. Nevertheless, the deviation involved here is de minimis, and, regardless of what other infirmities the *783plan may have, constitutional or otherwise, there is no violation of Art. I, §2 — the sole issue before us. It would, of course, be a different matter if appellees could demonstrate that New Jersey’s plan invidiously discriminated against a racial or political group. See White v. Regester, supra; Gaffney v. Cummings, supra, at 751-754; Whitcomb v. Chavis, 403 U. S. 124 (1971); Gomillion v. Lightfoot, 364 U. S. 339 (1960).
IV
Even if the Court’s view of the law were correct, its disposition of the case is not. At a minimum, the Court should vacate the decision of the District Court and remand for further consideration. As previously indicated, the Court finally recognizes today that considerations such as respecting political subdivisions and avoiding contests between incumbent Representatives might justify small population variances. Indeed, the Court indicates that “any number of consistently applied legislative policies” might do so. Ante, at 740. There is evidence in the record to suggest that the New Jersey Legislature was concerned with such considerations.16 The Court itself notes: “many of the problems that the New Jersey Legislature encountered in drawing districts with equal population stemmed from the decision . . . not to divide any municipalities between two congressional districts.” Ante, at 733, n. 5. But even if there were no evidence in the record, the State should be given a chance to defend its plan on this basis. Surely, the Court cannot rely on the fact that appellants have advanced only one justification for the plan’s population deviations — preserving the voting strength of racial minority groups. Relying on Kirkpatrick and White v. Weiser, supra, appellants no doubt concluded that other justifications were foreclosed and that the introduction of such proof would be futile.
By 1962, 35 out of 42 States had variances among their districts of over 100,000. Wesberry v. Sanders, 376 U. S. 1, 20-21 (1964) (Harlan, J. dissenting). The Court has recognized the significance of the fact that “enormous variations” in district size were at issue in the early legislative apportionment cases. Gaffney v. Cummings, 412 U. S. 735, 744, and n. 9 (1973).
District Courts have upheld or selected plans with similar deviations. See, e. g., Doulin v. White, 535 F. Supp. 450, 451 (ED Ark. 1982) (court ordered implementation of plan with 0.78% deviation despite alternative plan with deviation of 0.13%).
U. S. Bureau of the Census, Users’ Guide, 1980 Census of Population and Housing 100 (Mar. 1982). The National Academy of Sciences has estimated that the national undercount in the 1970 census was 2.5%. Panel on Decennial Census Plans, Counting the People in 1980: An Appraisal of Census Plans 2 (1978). One estimate is that the undercount error in the 1980 census is likely to be more than 2 million people nationwide, App. 103 (Dr. Trussel), and may be as high as 5 million. J. Passel, J. Siegel, & J. Robinson, Coverage of the National Population in the 1980 Census, by Age, Sex, and Race: Preliminary Estimates by Demographic Analysis (Nov. 1981) (Record Doc. No. 31).
The Court, after professing to “[a]ssum[e] for purposes of argument that each of [Dr. Trussel’s] statements is correct,” ante, at 735-736, proceeds in the following paragraph to denigrate his calculation as guesswork because the margin of statistical imprecision, i. e., the undercounting of persons, cannot be known precisely. The failure to quantify uncertainty exactly does not excuse pretending that it does not exist. When the question is whether the range of error is 1% or 2% or 2.5% and the deviation at hand is no larger than 0.6984%, the question is more academic than practical. Moreover, if a fixed benchmark were required, the margin of error officially recognized by the Census Bureau — last estimated at 2.3% — could easily be selected.
The Court also makes much of the fact that the precise amount of variation in undercounting among districts cannot be known with certainty. The relevant point, however, is that these district-to-district variances make it impossible to determine with statistical confidence whether opting for the plan with the smallest maximum deviation is ameliorating or aggravating actual equality of population among the districts. In addition, the count of individuals per district depends upon the Census Bureau’s selection of geographic boundaries by which to group data. “Data from the 1980 census have been compiled for congressional districts by equating component census geographic areas to each district and summing all data for areas coded to the district. Where the smallest census geographic area was split by a congressional district boundary, the census maps for the area were reviewed to determine in which district the majority of the population fell, and the entire area was coded to that district.” U. S. Bureau of Census, Congressional Districts of the 98th Congress A-l (1983) (preliminary draft). Thus, completely aside from undercounting effects, it is obvious that even absolute numerical equality between the census figures for congressional districts does not reflect districts of equal size.
*771Finally, the Court dismisses the entire concept of statistical error with the sophistic comment that “[e]ven if one cannot say with certainty that one district is larger than another merely because it has a higher census count, one can say with certainty that the district with a larger census count is more likely to be larger than the other district than it is to be smaller or the same size.” Ante, at 738. The degree of that certainty, however, is speculative. The relevant consideration is not whether District Four is larger than District Six, but how much larger, and, how much less larger under the selected plan vis-á-vis an alternative plan. Moreover, variable undercounting and differences between census units and district lines may result in other districts having higher maximum deviations.
The general point is that when the numbers become so small, it makes no sense to concentrate on ever finer gradations when one cannot even be certain whether doing so increases or decreases actual population variances.
In New Jersey, for example, population growth during the 1970’s enlarged some districts by up to 26%, while other congressional districts lost up to 8.7% of their 1970 population. U. S. Bureau of Census, Congressional Districts of the 98th Congress 32-3 (1983). See also Gaffney v. Cummings, 412 U. S., at 746, n. 11.
Justice Stevens makes the same point.
“Given the birth rate, the mortality rate, the transient character of modern society, and the acknowledged errors in the census, we all know that such differences may vanish between the date of the census and the date of the next election. Absolute population equality is impossible to achieve.” Ante, at 752 (concurring opinion).
In New Jersey, for example, the population 18 years old and over differs significantly among the congressional districts. In 1978, District 10 had but 282,000 such individuals, while District 2 had 429,000. U. S. Bureau of Census, State and Metropolitan Area Data Book 549 (1979). See also Gaffney v. Cummings, supra, at 747, n. 18.
Throughout the Nation, approximately 71% of the voting age population registers to vote. U. S. Bureau of Census, State and Metropolitan Area Data Book 567 (1982).
As a result of all these factors, as well as the failure of many registered voters to cast ballots, the weight of a citizen’s vote in one district is inevitably different from that in others. For example, the total number of votes cast in the 1982 New Jersey congressional races differed significantly between districts, ranging from 92,852 in District 10 to 186,879 in District 9. 41 Congressional Quarterly 391 (1983).
States with larger deviations are Indiana (2.96%); Alabama (2.45%); Tennessee (2.40%); Georgia (2.00%); Virginia (1.81%); North Carolina (1.76%); New York (1.64%); Kentucky (1.39%); Washington (1.30%); Massachusetts (1.09%); New Mexico (0.87%); Arkansas (0.78%). States with similar maximum deviations are Ohio (0.68%); Nevada (0.60%); Oklahoma (0.58%); West Virginia (0.49%). Council of State Governments & National Conference of State Legislatures, 1 Reapportionment Information Update 6-7 (Nov. 12, 1982).
The Court’s language suggests that not only must the maximum variance in a plan be supported, but that also every deviation from absolute equality must be so justified. Ante, at 740. Consider the staggering nature of the burden imposed: Each population difference between any two districts in a State must be justified, apparently even if none of the plans before the legislature or commission would have reduced the difference. See n. 11, infra.
Even by choosing the plan with the smallest deviation, a legislature or commission cannot be assured of avoiding constitutional challenge. In this case the Court does not find that the 0.6984% deviation was avoidable because there were other plans before the New Jersey Legislature with smaller maximum variations. Nor does the Court counter appellants’ position, supported by evidence in the record, that these alternative plans had other disqualifying faults. Instead, the Court tries its own hand at redistricting New Jersey and concludes that by moving around 13 New Jersey subdivisions, the maximum deviation could be reduced to 0.449%. Ante, at 739-740, n. 10. The message for state legislatures is clear: it is not enough that the chosen plan be superior to any actual plans introduced as' alternatives, the plan must also be better than any conceivable alternative a federal judge can devise.
Unlike population deviations, political gerrymandering does not lend itself to arithmetic proof. Nevertheless, after reviewing the recent redistricting throughout the country, one commentator offered the following assessment:
“The nobly aimed ‘one-man, one-vote’ principle is coming into increasing use as a weapon for state legislators bent on partisan gerrymandering. From California to New Jersey and points in between, Republicans and Democrats alike are justifying highly partisan remaps by demonstrating respect for the 1964 Supreme Court mandate that population of congressional districts within states must be made as equal as possible. Meanwhile, other interests at stake in redistricting — such as the preservation of community boundaries and the grouping of constituencies with similar concerns — are being brushed aside .... The emphasis on one-man, one-vote not only permits gerrymandering, it encourages it. In many states it is *777impossible to approach population equality without crossing city, county and township lines. Once the legislature recognizes that move must be made, it is only a short step further to the drawing of a line that dances jaggedly through every region of the state. Local interests, informed that it is no longer legally permissible to draw a whole-county congressional map in most states, are far less likely to object than they were in the past.... The court’s decision to reject a tiny deviation in favor of an even smaller one may further encourage the hairsplitting numbers game that has given rise to partisan gerrymanders all over the country.” Congressional Quarterly, Inc., State Politics and Redistricting 1-2 (1982).
See also Engstrom, The Supreme Court and Equipopulous Gerrymandering: A Remaining Obstacle in the Quest for Fair and Effective Representation, 1976 Ariz. State L. J. 277, 278 (“Not only has the Court failed to develop effective checks on the practice of gerrymandering, but in pursuing the goal of population equality to a point of satiety it has actually facilitated that practice”); Baker, One Man, One Vote, and “Political Fairness,” 23 Emory L. J. 701, 710 (1974) (hereafter Baker) (“Priority was typically given to miniscule population variations at the expense of any recognition of political subdivisions. Charges of partisan gerrymandering were more widespread than in past decades for two major reasons: the extent of redistricting activity among all fifty states, and the lack of emphasis on former norms of compactness and adherence to local boundary lines”).
In the eyes of some commentators, the experience of New York in the aftermath of Wells v. Rockefeller is instructive.
“Subsequent congressional districting in New York became a possible prototype for the ‘equal-population gerrymander.’ Whereas the former district pattern nullified by the Supreme Court had been the result of bipartisan compromise with each major party controlling one house, by 1970 the Republicans held both legislative houses as well as the governorship. The assistant counsel to the senate majority leader (and chief coordinator of the redistricting) candidly remarked: ‘The Supreme Court is just making gerrymandering easier than it used to be.’ Not only was New York City subjected to major cartographic surgery, but upstate cities were also fragmented, with portions being joined to suburban and rural areas in an attempt to dilute concentrations of Democrats.” Baker, at 712-713. Yet, under the new plan, no district deviated by more than than 490 persons from the average, and the configuration of district boundaries revealed generally compact and contiguous contours. Baker, Gerrymander*778ing: Privileged Sanctuary or Next Judicial Target?, in Reapportionment in the 1970s, p. 138 (N. Polsby ed. 1971). Ironically, David Wells, the plaintiff who successfully challenged the former district pattern, returned to federal court in February 1970 to ask if the old plan could be restored. See Dixon, “One Man, One Vote — What Happens Next?,” 60 Nat. Civic Rev. 259, 265 (1971).
See also Mahan v. Howell, 410 U. S. 315, 341 (1973) (Brennan, J., concurring in part and dissenting in part) (“What our decisions have made clear is that certain state interests that are pertinent to legislative reapportionment can have no possible relevance to congressional districting. Thus, the need to preserve the integrity of political subdivisions as political subdivisions may, in some instances, justify small variations in the population of districts from which state legislators are elected. But that interest can hardly be asserted in justification of malapportioned congressional districts. Kirkpatrick v. Preisler, supra”).
As the law has developed, our congressional cases are rooted in Art I, § 2, of the Constitution while our legislative cases rely upon the Equal Protection Clause of the Fourteenth Amendment. I am not aware, however, of anything in the respective provisions which justifies, let alone requires, the difference in treatment that has emerged between the two lines of decisions. Our early cases were frequently cross-cited, and the formulation “as nearly of equal population as is practicable” appears in Reynolds v. Sims, 377 U. S., at 589, as well as in Wesberry v. Sanders, 376 U. S., at 7-8. The differing paths the cases have taken since Kirkpatrick must result from that decision’s rejection of the legitimacy of considering nonpopu-lation factors in congressional redistricting. See Mahan v. Howell, 410 U. S., at 341 (Brennan, J., concurring in part and dissenting in part). With today’s long-awaited overruling of that holding in Kirkpatrick, any remaining justification disappears for such a marked difference in our approach to congressional and legislative reapportionment.
Experience in the legislative apportionment field following our allowance of a range of de minimis variance is convincing proof that we need not fear that the goal of equal population in the districts will receive less than its due. Justice Brennan’s prediction that tolerating de minimis population variances would “jeopardize the very substantial gains” made in equalizing legislative districts, White v. Regester, 412 U. S. 755, 781 (1973) (concurring in part and dissenting in part), has not been proved, and, indeed, the prediction is refuted by an analysis of the legislative redistricting undertaken after the 1980 census. See Council of State Governments & National Conference of State Legislatures, 1 Reapportionment Information Update 6 (Nov. 12, 1982).
See, e. g., Feldman Deposition, at 91-94 (Record Doc. No. 39) (concern with fairness to incumbents); Jackman Deposition, at 91-92 (Record Doc. No. 40) (concern with preserving political subdivisions).