delivered the opinion of the Court.
This case presents the question whether a party’s post-filing change in citizenship can cure a lack of subject-matter jurisdiction that existed at the time of filing in an action premised upon diversity of citizenship. See 28 U. S. C. § 1332.
I
Respondent Atlas Global Group, L. R, is a limited partnership created under Texas law. In November 1997, Atlas filed a state-law suit against petitioner Grupo Dataflux, a Mexican corporation, in the United States District Court for the Southern District of Texas. The complaint contained claims for breach of contract and in quantum meruit, seeking over $1.3 million in damages. It alleged that “[fjederal jurisdiction is proper based upon diversity jurisdiction pursuant to 28 U. S. C. § 1332(a), as this suit is between a Texas citizen [Atlas] and a citizen or subject of Mexico [Grupo Da-*569taflux].”1 App. 19a (Complaint ¶3). Pretrial motions and discovery consumed almost three years. In October 2000, the parties consented to a jury trial presided over by a Magistrate Judge. On October 27, after a 6-day trial, the jury returned a verdict in favor of Atlas awarding $750,000 in damages.
On November 18, before entry of the judgment, Dataflux filed a motion to dismiss for lack of subject-matter jurisdiction because the parties were not diverse at the time the complaint was filed. See Fed. Rules Civ. Proe. 12(b)(1), (h)(3). The Magistrate Judge granted the motion. The dismissal was based upon the accepted rule that, as a partnership, Atlas is a citizen of each State or foreign country of which any of its partners is a citizen. See Carden v. Arkoma Associates, 494 U. S. 185, 192-195 (1990). Because Atlas had two partners who were Mexican citizens at the time of filing, the partnership was a Mexican citizen. (It was also a citizen of Delaware and Texas based on the citizenship of its other partners.) And because the defendant, Dataflux, was a Mexican corporation, aliens were on both sides of the case, and the requisite diversity was therefore absent. See Mossman v. Higginson, 4 Dall. 12, 14 (1800).
On appeal, Atlas did not dispute the finding of no diversity at the time of filing. It urged the Court of Appeals to disregard this failure and reverse dismissal because the Mexican partners had left the partnership in a transaction consummated the month before trial began. Atlas argued that, since diversity existed when the jury rendered its verdict, dismissal was inappropriate. The Fifth Circuit agreed. 312 F. 3d 168, 174 (2002). It acknowledged the general rule that, for purposes of determining the existence of diversity *570jurisdiction, the citizenship of the parties is to be determined with reference to the facts as they existed at the time of filing. Id,., at 170. However, relying on our decision in Caterpillar Inc. v. Lewis, 519 U. S. 61 (1996), it held that the conclusiveness of citizenship at the time of filing was subject to exception when the following conditions are satisfied:
“(1) [A]n action is filed or removed when constitutional and/or statutory jurisdictional requirements are not met, (2) neither the parties nor the judge raise the error until after a jury verdict has been rendered, or a disposi-tive ruling has been made by the court, and (3) before the verdict is rendered, or ruling is issued, the jurisdictional defect is cured.” 312 F. 3d, at 174.
The opinion strictly limited the exception as follows: “If at any point prior to the verdict or ruling, the issue is raised, the court must apply the general rule and dismiss regardless of subsequent changes in citizenship.” Ibid.
The jurisdictional error in the present case not having been identified until after the jury returned its verdict; and the postfiling change in the composition of the partnership having (in the Court’s view) cured the jurisdictional defect; the Court reversed and remanded with instructions to the District Court to enter judgment in favor of Atlas. Ibid. We granted certiorari. 540 U. S. 944 (2008).
II
It has long been the case that “the jurisdiction of the court depends upon the state of things at the time of the action brought.” Mollan v. Torrance, 9 Wheat. 537, 539 (1824). This time-of-filing rule is hornbook law (quite literally2) *571taught to first-year law students in any basic course on federal civil procedure. It measures all challenges to subject-matter jurisdiction premised upon diversity of citizenship against the state of facts that existed at the time of filing — whether the challenge be brought shortly after filing, after the trial, or even for the first time on appeal. (Challenges to subject-matter jurisdiction can of course be raised at any time prior to final judgment. See Capron v. Van Noorden, 2 Cranch 126 (1804).)
We have adhered to the time-of-filing rule regardless of the costs it imposes. For example, in Anderson v. Watt, 138 U. S. 694 (1891), two executors of an estate, claiming to be New York citizens, had brought a diversity-based suit in federal court against defendants alleged to be Florida citizens. When it later developed that two of the defendants were New York citizens, the plaintiffs sought to save jurisdiction by revoking the letters testamentary for one executor and alleging that the remaining executor was in fact a British citizen. The Court rejected this attempted postfiling salvage operation, because at the time of filing the executors included a New Yorker. Id., at 708. It dismissed the case for want of jurisdiction, even though the case had been filed about 5V2 years earlier, the trial court had entered a decree ordering land to be sold 4 years earlier, the sale had been made, exceptions had been filed and overruled, and the case had come to the Court on appeal from the order confirming the land sale. Id., at 698. Writing for the Court, Chief Justice Fuller adhered to the principle set forth in Conolly v. Taylor, 2 Pet. 556, 565 (1829), that “jurisdiction depending on the condition of the party is governed by that condition, as it was at the commencement of the suit.” “[Jurisdiction,” he reasoned, “could no more be given ... by the amendment than if a citizen of Florida had sued another in *572that court and subsequently sought to give it jurisdiction by removing from the State.” 138 U. S., at 708.3
It is uncontested that application of the time-of-filing rule to this case would require dismissal, but Atlas contends that this Court “should accept the very limited exception created by the Fifth Circuit to the time-of-filing principle.” Brief for Respondents 2. The Fifth Circuit and Atlas rely on our statement in Caterpillar, supra, at 75, that “[o]nce a diversity case has been tried in federal court . . . considerations of finality, efficiency, and economy become overwhelming.” This statement unquestionably provided the ratio decidendi in Caterpillar, but it did not augur a new approach to deciding whether a jurisdictional defect has been cured.
Caterpillar broke no new ground, because the jurisdictional defect it addressed had been cured by the dismissal of the party that had destroyed diversity. That method of curing a jurisdictional defect had long been an exception to the time-of-filing rule. “[T]he question always is, or should be, when objection is taken to the jurisdiction of the court by reason of the citizenship of some of the parties, whether ... they are indispensable parties, for if their interests are sev-erable and a decree without prejudice to their rights may be made, the jurisdiction of the court should be retained and the suit dismissed as to them.” Horn v. Lockhart, 17 Wall. 570, 579 (1873). Federal Rule of Civil Procedure 21 provides that “[p]arties may be dropped or added by order of the court on motion of any party or of its own initiative at any stage of the action and on such terms as are just.” By now, “it is *573well settled that Rule 21 invests district courts with authority to allow a dispensable nondiverse party to be dropped at any time, even after judgment has been rendered.” Newman-Green, Inc. v. Alfonzo-Larrain, 490 U. S. 826, 832 (1989). Indeed, the Court held in Newman-Green that courts of appeals also have the authority to cure a jurisdictional defect by dismissing a dispensable nondiverse party. Id., at 837.
Caterpillar involved an unremarkable application of this established exception. Complete diversity had been lacking at the time of removal to federal court, because one of the plaintiffs shared Kentucky citizenship with one of the defendants. Almost three years after the District Court denied a motion to remand, but before trial, the diversity-destroying defendant settled out of the case and was dismissed. The case proceeded to a 6-day jury trial, resulting in judgment for the defendant, Caterpillar, against Lewis. This Court unanimously held that the lack of complete diversity at the time of removal did not require dismissal of the case.
The sum of Caterpillar’s jurisdictional analysis was an approving acknowledgment of Lewis’s admission that there was “complete diversity, and therefore federal subject-matter jurisdiction, at the time of trial and judgment.” 519 U. S., at 73. The failure to explain why this solved the problem was not an oversight, because there was nothing novel to explain. The postsettlement dismissal of the diversity-destroying defendant cured the jurisdictional defect just as the dismissal of the diversity-destroying party had done in Newman-Green. In both cases, the less-than-complete diversity which had subsisted throughout the action had been converted to complete diversity between the remaining parties to the final judgment. See also Horn, supra, at 579.
While recognizing that Caterpillar is “technically” distinguishable because the defect was cured by the dismissal of a diversity-destroying party, the Fifth Circuit reasoned that *574“this factor was not at the heart of the Supreme Court’s analysis____” 312 F. 3d, at 172-173. The crux of the analysis, according to the Fifth Circuit, was Caterpillar’s statement that “[o]nce a diversity case has been tried in federal court . . . considerations of finality, efficiency, and economy become overwhelming.” 519 U. S., at 75. This was indeed the crux of analysis in Caterpillar, but analysis of a different issue. It related not to cure of the jurisdictional defect, but to cure of a statutory defect, namely, failure to comply with the requirement of the removal statute, 28 U. S. C. § 1441(a), that there be complete diversity at the time of removal.4 The argument to which the statement was directed took as its starting point that subject-matter jurisdiction had been satisfied: “ultimate satisfaction of the subject-matter jurisdiction requirement ought not swallow up antecedent statutory violations.” 519 U. S., at 74 (emphasis added). The resulting holding of Caterpillar, therefore, is only that a statutory defect — “Caterpillar’s failure to meet the § 1441(a) requirement that the case be fit for federal adjudication at the time the removal petition is filed,” id., at 73 — did not require dismissal once there was no longer any jurisdictional defect.
III
To our knowledge, the Court has never approved a deviation from the rule articulated by Chief Justice Marshall in 1829 that “[wjhere there is no change of party, a jurisdiction depending on the condition of the party is governed by that condition, as it was at the commencement of the suit.” Conolly, 2 Pet., at 565 (emphasis added). Unless the Court is to manufacture a brand-new exception to the time-of-filing rule, *575dismissal for lack of subject-matter jurisdiction is the only option available in this case. The purported cure arose not from a change in the parties to the action, but from a change in the citizenship of a continuing party. Withdrawal of the Mexican partners from Atlas did not change the fact that Atlas, the single artificial entity created under Texas law, remained a party to the action. True, the composition of the partnership, and consequently its citizenship, changed. But allowing a citizenship change to cure the jurisdictional defect that existed at the time of filing would contravene the principle articulated by Chief Justice Marshall in Conolly5 We decline to do today what the Court has refused to do for the past 175 years.
Apart from breaking with our longstanding precedent, holding that “finality, efficiency, and judicial economy” can justify suspension of the time-of-filing rule would create an exception of indeterminate scope. The Court of Appeals sought to cabin the exception with the statement that “[i]f at any point prior to the verdict or ruling, the [absence of diversity at the time of filing] is raised, the court must apply the general rule and dismiss regardless of subsequent *576changes in citizenship,” 312 F. 3d, at 174. This limitation is unsound in principle and certain to be ignored in practice.
It is unsound in principle because there is no basis in reason or logic to dismiss preverdict if in fact the change in citizenship has eliminated the jurisdictional defect. Either the court has jurisdiction at the time the defect is identified (because the parties are diverse at that time) or it does not (because the postfiling citizenship change is irrelevant). If the former, then dismissal is inappropriate; if the latter, then retention of jurisdiction postverdict is inappropriate.
Only two escapes from this dilemma come to mind, neither of which is satisfactory. First, one might say that it is not any change in party citizenship that cures the jurisdictional defect, but only a change that remains unnoticed until the end of trial. That is not so much a logical explanation as a restatement of the illogic that produces the dilemma. There is no conceivable reason why the jurisdictional deficiency which continues despite the citizenship change should suddenly disappear upon the rendering of a verdict. Second, one might say that there never was a cure, but that the party who failed to object before the end of trial forfeited his objection. This is logical enough, but comes up against the established principle, reaffirmed earlier this Term, that “a court’s subject-matter jurisdiction cannot be expanded to account for the parties’ litigation conduct.” Kontrick v. Ryan, 540 U. S. 443, 456 (2004). “A litigant generally may raise a court’s lack of subject-matter jurisdiction at any time in the same civil action, even initially at the highest appellate instance.” Id., at 455. Because the Fifth Circuit’s attempted limitation upon its new exception makes a casualty either of logic or of this Court’s jurisprudence, there is no principled way to defend it.
And principled or not, the Fifth Circuit’s artificial limitation is sure to be discarded in practice. Only 8% of diversity cases concluded in 2003 actually went to trial, and the median time from filing to trial disposition was nearly two years. *577See Administrative Office of the United States Courts, Statistics on Diversity Filings and Terminations in District Courts for Calendar Year 2008 (on file with the Clerk of Court). In such a litigation environment, an approach to jurisdiction that focuses on efficiency and judicial economy cannot possibly be held to the line drawn by the Court of Appeals. As Judge Garza observed in his dissent:
“[TJhere is no difference in efficiency terms between the jury verdict and, for example, the moment at which the jury retires. Nor, for that matter, is there a large difference between the verdict and mid-way through the trial... . Indeed, in complicated cases requiring a great deal of discovery, the parties and the court often expend tremendous resources long before the case goes to trial.” 312 F. 3d, at 177.
IV
The dissenting opinion rests on two principal propositions: (1) the jurisdictional defect in this case was cured by a change in the composition of the partnership; and (2) refusing to recognize an exception to the time-of-filing rule in this case wastes judicial resources, while creating an exception does not. We discuss each in turn.
A
Unlike the dissent, our opinion does not turn on whether the jurisdictional defect here contained at least “minimal diversity.”6 Regardless of how one characterizes the ac*578knowledged jurisdictional defect, it was never cured. The only two ways in which one could conclude that it had been cured would be either (1) to acknowledge that a party’s post-filing change of citizenship can cure a time-of-filing jurisdictional defect, or (2) to treat a change in the composition of a partnership like a change in the parties to the action. The Court has never, to our knowledge, done the former; and not even the dissent suggests that it ought to do so in this case.7 The dissent diverges from our analysis by adopting the latter approach, stating that “this case seems ... indistinguishable from one in which there is a change in the parties to the action.” Post, at 591 (internal quotation marks omitted).
This equation of a dropped partner with a dropped party is flatly inconsistent with Carden. The dissent in Carden sought to apply a “real party to the controversy” approach to determine which partners counted for purposes of jurisdictional analysis. The Carden majority rejected that ap*579proach, reasoning that “[t]he question presented today is not which of various parties before the Court should be considered for purposes of determining whether there is complete diversity of citizenship .... There are not . . . multiple respondents before the Court, but only one: the artificial entity called Arkoma Associates, a limited partnership.” 494 U. S., at 188, n. 1. Today’s dissent counters that “[w]hile a partnership may be characterized as a single artificial entity, a district court determining whether diversity jurisdiction exists looks to the citizenship of the several persons composing [the entity].” Post, at 591, n. 8 (internal quotation marks and citations omitted). It is true that the court “looks to” the citizenship of the several persons composing the entity, but it does so for the purpose of determining the citizenship of the entity that is a party, not to determine which citizens who compose the entity are to be treated as parties. See Carden, 494 U. S., at 188, n. 1 (“[W]hat we must decide is the . . . question of how the citizenship of that single artificial entity is to be determined”); id., at 195 (“[W]e reject the contention that to determine, for diversity purposes, the citizenship of an artificial entity, the court may consult the citizenship of less than all of the entity’s members”).8
There was from the beginning of this action a single plaintiff (Atlas), which, under Carden, was not diverse from the sole defendant (Dataflux). Thus, this case fails to present “two adverse parties [who] are not co-citizens.” State Farm Fire & Casualty Co. v. Tashire, 386 U. S. 523, 531 (1967). Contrary to the dissent’s characterization, then, this is not a *580case like Caterpillar or Newman-Green in which “party lineup changes . . . simply trimmed the litigation down to an ever-present core that met the statutory requirement.” Post, at 591. Rather, this is a case in which a single party changed its citizenship by changing its internal composition.
The incompatibility with prior law of the dissent’s attempt to treat a change in partners like a change in parties is revealed by a curious anomaly: It would produce a case unlike every other ease in which dropping a party has cured a jurisdictional defect, in that no judicial action (such as granting a motion to dismiss) was necessary to get the jurisdictional spoilers out of the case. Indeed, judicial action to that end was not even possible: The court could hardly have “dismissed” the partners from the partnership to save jurisdiction.9
B
We now turn from consideration of the conceptual difficulties with the dissent’s disposition to consideration of its practical consequences. The time-of-filing rule is what it is precisely because the facts determining jurisdiction are subject to change, and because constant litigation in response to that change would be wasteful. The dissent would have it that the time-of-filing rule applies to establish that a court has jurisdiction (and to protect that jurisdiction from later destruction), but does not apply to establish that a court lacks jurisdiction (and to prevent postfiling changes that perfect jurisdiction). Post, at 583-584. But whether destruction or perfection of jurisdiction is at issue, the policy goal *581of minimizing litigation over jurisdiction is thwarted whenever a new exception to the time-of-filing rule is announced, arousing hopes of further new exceptions in the future. Cf. Dretke v. Haley, ante, at 394-395 (recognizing that the creation of exceptions to judge-made procedural rules will enmesh the federal courts in litigation testing the boundaries of each new exception). That litigation-fostering effect would be particularly strong for a new exception derived from such an expandable concept as the “efficiency” rationale relied upon by the dissent.
The dissent argues that it is essential to uphold jurisdiction in this and similar cases because dismissal followed by refiling condemns the parties to “an almost certain replay of the case, with, in all likelihood, the same ultimate outcome.” Post, at 595. But if the parties expect “the same ultimate outcome,” they will not waste time and resources slogging through a new trial. They will settle, with the jury’s prior verdict supplying a range for the award. Indeed, settlement instead of retrial will probably occur even if the parties do not expect the same ultimate outcome. When the stakes remain the same and the players have been shown each other’s cards, they will not likely play the hand all the way through just for the sake of the game. And finally, even if the parties run the case through complete “relitigation in the very same District Court in which it was first filed in 1997,” post, at 598, the “waste” will not be great. Having been through three years of discovery and pretrial motions in the current case, the parties would most likely proceed promptly to trial.
Looked at in its overall effect, and not merely in its application to the sunk costs of the present case, it is the dissent’s proposed rule that is wasteful. Absent uncertainty about jurisdiction (which the dissent’s readiness to change settled law would preserve for the future), the obvious course, for a litigant whose suit was dismissed as Atlas’s was, would have been immediately to file a new action. That is in fact what *582Atlas did, though it later dismissed the new case without prejudice. Had that second suit been pursued instead of this one, there is little doubt that the dispute would have been resolved on the merits by now. Putting aside the time that has passed between the Fifth Circuit’s decision and today, there were two years of wasted time between dismissal of the action and the Fifth Circuit’s reversal of that dismissal — time that the parties could have spent litigating the merits (or engaging in serious settlement talks) instead of litigating jurisdiction.
Atlas and Dataflux have thus far litigated this case for more than 6V2 years, including BV2 years over a conceded jurisdictional defect. Compared with the one month it took the Magistrate Judge to apply the time-of-filing rule and Carden when the jurisdictional problem was brought to her attention, this waste counsels strongly against any course that would impair the certainty of our jurisdictional rules and thereby encourage similar jurisdictional litigation.
We decline to endorse a new exception to a time-of-filing rule that has a pedigree of almost two centuries. Uncertainty regarding the question of jurisdiction is particularly undesirable, and collateral litigation on the point particularly wasteful. The stability provided by our time-tested rule weighs heavily against the approval of any new deviation. The judgment of the Fifth Circuit is reversed.
It is so ordered.
Title 28 U. S. C. § 1332(a)(2) provides:
“The district courts shall have original jurisdiction of all civil actions where the matter in controversy exceeds the sum of $75,000, exclusive of interest and costs, and is between ...
“citizens of a State and citizens or subjects of a foreign state.”
See, e. g., J. Friedenthal, M. Kane, & A. Miller, Civil Procedure 27 (3d ed. 1999); C. Wright & M. Kane, Law of Federal Courts 173 (6th ed. 2002). See also 13B C. Wright, A. Miller, & E. Cooper, Federal Practice and Procedure §3608, p. 452 (1984).
The dissent asserts that Anderson is “not altogether in tune with Caterpillar and Newman-Green,” post, at 591, n. 7 (opinion of Ginsburg, J.), but the cases can easily be harmonized. Anderson did not, as the dissent suggests, refuse to give diversity-perfecting effect to the dismissal of an independent severable party; it refused to give that effect to the alteration of a coexecutorship into a lone executorship — much as we decline to give diversity-perfecting effect to the alteration of a partnership with diversity-destroying partners into a partnership without diversity-destroying partners.
Title 28 U. S. C. § 1441(a) provides, in relevant part:
“Except as otherwise expressly provided by Act of Congress, any civil action brought in a State court of which the district courts of the United States have original jurisdiction, may be removed by the defendant or the defendants, to the district court of the United States for the district and division embracing the place where such action is pending.”
The dissent acknowledges that “[t]he Court has long applied [Chief Justice] Marshall’s time-of-filing rule categorically to postfiling changes that otherwise would destroy diversity jurisdiction,” post, at 583-584, but asserts that “[i]n contrast, the Court has not adhered to a similarly steady rule for postfiling changes in the party lineup, alterations that perfect previously defective statutory subject-matter jurisdiction,” post, at 584. The authorities relied upon by the dissent do not call into question the particular aspect of the time-of-filing rule that is at issue in this case— the principle (quoted in text) that “[w]here there is no change of party, a jurisdiction depending on the condition of the party is governed by that condition, as it was at the commencement of the suit.” Conolly, 2 Pet., at 565 (emphasis added). The dissent identifies five cases in which the Court permitted a postfiling change to cure a jurisdictional defect. Post, at 584. Every one of them involved a change of party. The dissent does not identify a single case in which the Court held that a single party’s postfiling change of citizenship cured a previously existing jurisdictional defect.
The answer to the “minimal diversity” question is not as straightforward as the dissent’s analysis suggests. We understand “minimal diversity” to mean the existence of at least one party who is diverse in citizenship from one party on the other side of the case, even though the extraconstitutional “complete diversity” required by our cases is lacking. It is possible, though far from clear, that one can have opposing parties in a two-party case who are cocitizens, and yet have minimal Article III jurisdiction because of the multiple citizenship of one of the parties. Although the Court has previously said that minimal diversity requires “two *578adverse parties [who] are not co-citizens,” State Farm Fire & Casualty Co. v. Tashire, 386 U. S. 523, 531 (1967), the Court did not have before it a multiple-citizenship situation.
The dissent contends that the existence of minimal diversity was clear because the rule of Carden v. Arkoma Associates, 494 U. S. 185 (1990), is not required by the Constitution. Post, at 588-590. But neither is the rule that a corporation is “a citizen of any State by which it has been incorporated and of the State where it has its principal place of business.” 28 U. S. C. § 1332(c)(1). We do not understand the inquiry into minimal diversity to proceed by hypothetically rewriting, to whatever the Constitution might allow that would support Article III jurisdiction in the particular case, all laws bearing upon the diversity question. Whether the Constitution requires it or not, Carden is the subconstitutional rule by which we determine the citizenship of a partnership — and in this case it leads to the conclusion that there were no opposing parties who were not cocitizens.
The dissent appears to leave open the possibility that this line could be crossed in a future case, contrasting Caterpillar Inc. v. Lewis, 519 U. S. 61 (1996), not with all cases involving a party’s change of citizenship, but with the polar extreme of “a plaintiff who moves to another State to create diversity not even minimally present when the complaint was filed,” post, at 590.
These statements from Carden rebut the dissent’s assertion that “an association whose citizenship, for diversity purposes, is determined by aggregating the citizenships of each of its members” could “[w]ith equal plausibility ... be characterized as an ‘aggregation’ composed of its members, or an ‘entity’ comprising its members.” Post, at 590, n. 6. We think it evident that Carden decisively adopted an understanding of the limited partnership as an “entity,” rather than an “aggregation,” for purposes of diversity jurisdiction. See 494 U. S., at 188, n. 1.
An additional anomaly, under the particular facts of the present ease, is that the two individual Mexican partners, whom the dissent treats like parties for purposes of enabling their withdrawal to perfect jurisdiction, were brought into the litigation personally by the court’s granting of Dataflux’s motion to add them as parties for purposes of Dataflux’s counterclaim. The motion was made and granted under Federal Rule of Civil Procedure 13(h), which applies only to “[plersons other than those made parties to the original action.” (Emphasis added.)