Is Voter Fraud a Real Problem?
How much voter fraud is on record, set against the ballots cast; which kinds are rarest, including the one voter ID is aimed at; what the 2020 lawsuits and the investigations since found; and the strongest case that the record misses a great deal.
Jump to the verdict ↓A small one. Voter fraud happens and is prosecuted, but rarely. The Heritage Foundation’s database, which says itself that it is not complete, lists 1,619 cases from 1982 to 2025, about 7 for every 10 million ballots cast in federal general elections alone. Fraud has forced at least one election to be re-run: North Carolina’s 9th congressional district in 2018, after an absentee-ballot scheme. It has not come near a presidential result. Reversing 2020 meant wiping out Biden’s margins in Arizona, Georgia and Wisconsin, 42,918 votes together; Heritage lists 17 cases from that election in those states, and the one win among the 64 lawsuits that followed was too small to matter.
The Scale of American Elections
The Heritage Foundation, a conservative think tank, keeps an Election Fraud Database (electionfraud.heritage.org). At its last update, on December 12, 2025, it listed 1,619 cases dated from 1982 to 2025, each ending in a criminal conviction (about 1,380 of them), a diversion program that clears the case once the defendant completes it, a civil penalty, or a finding of fraud by a court or an official body. A case can be one person's single vote or a scheme involving many ballots. Heritage calls the list "a sampling" that "does not capture all cases", and says that "nobody really knows the extent of election fraud."
In the 22 federal general elections from 1982 to 2024, Americans cast about 2.26 billion ballots. Heritage's 1,619 cases come to about 7 for every 10 million of those ballots, or 0.00007%. That rate is generous to the fraud claim, because Heritage's cases also come from primaries and from state and local elections, whose ballots are not in the total.
How to read it. Each bar is one November federal election, 1982 at the left to 2024 at the right; presidential years alternate with lower midterms. Heights are ballots cast, in millions, from zero; the series gives no national count of ballots for 1982 to 1996, so those bars are votes for the highest office on the ballot, slightly lower. Values are in the table below. Source: the University of Florida Election Lab's national turnout series, from state election reports.
| Year | Ballots | Year | Ballots |
|---|---|---|---|
| 1982 | 67.6 | 1984 | 92.7 |
| 1986 | 65.0 | 1988 | 91.6 |
| 1990 | 67.9 | 1992 | 104.4 |
| 1994 | 75.1 | 1996 | 96.3 |
| 1998 | 74.8 | 2000 | 107.4 |
| 2002 | 80.3 | 2004 | 123.5 |
| 2006 | 85.8 | 2008 | 132.6 |
| 2010 | 90.9 | 2012 | 130.2 |
| 2014 | 83.2 | 2016 | 138.7 |
| 2018 | 118.6 | 2020 | 159.7 |
| 2022 | 112.0 | 2024 | 156.8 |
| All 22 elections | 2,255.1 | ||
Fraud changes a result only if it wipes out the winning margin. In 2020 Joe Biden carried the three closest states, Georgia, Arizona and Wisconsin, by 11,779, 10,457 and 20,682 votes, and Pennsylvania, the fourth closest by share of the vote, by 80,555 (certified results). Without the first three he would have had 269 electoral votes, one short of a majority, so 42,918 votes of margin in the right three states was the least that fraud would have had to wipe out; all four states together come to 123,473. Heritage's database holds 1,619 cases from every kind of election since 1982, and 17 from the 2020 general election in those three states.
None of this shows that no fraud went undetected. It shows the size an undetected scheme would have needed.
Fraud by Category
The kinds of fraud in Heritage's cases are far from equally common. The kind voter ID laws are built to stop, someone voting in another person's name at the polls, is among the rarest: 34 of the 1,619 cases name it, against 387 that involve absentee ballots.
How to read it. Each bar is one kind of fraud; its length is the number of Heritage's cases naming it, on an axis from zero. A case can name more than one kind, so the bars add to more than 1,619. Values are in the table below. Source: Heritage's database as of December 12, 2025.
| Kind of fraud | Cases |
|---|---|
| Absentee ballots | 387 |
| False registration | 386 |
| Felon voting | 234 |
| Vote buying | 208 |
| Duplicate voting | 177 |
| Petition fraud | 113 |
| Noncitizens | 98 |
| Impersonation at the polls | 34 |
Justin Levitt, a law professor, tracked every specific, credible allegation that someone had voted as another person at the polls and found about 31 incidents from 2000 to 2014, some involving more than one ballot, in years when general and primary elections alone drew more than a billion ballots (Washington Post, August 2014). He expected some to be debunked. The Justice Department made election crimes a priority for all its offices in October 2002; by July 2008 it had charged more than 140 people and convicted more than 100, and the New York Times reported in 2007, from court records and interviews, that the drive had turned up virtually no evidence of any organized effort to skew federal elections. Lorraine Minnite, reviewing the Department's own list of cases to September 2005, found that the 24 people convicted of illegal voting were 14 noncitizens, 5 still under state supervision for felony convictions and 5 who voted in two states. None had voted in someone else's name.
Absentee ballots are the most common kind, yet the Brookings Institution, counting Heritage's cases for the general elections of 2016 to 2022, put mail fraud at 0.000043% of mail ballots, about 4 in every 10 million.
| Count | Found | What happened |
|---|---|---|
| Heritage's database, 2000 to 2024 | 98 cases | Convictions and other findings |
| Georgia citizenship audit, 2026 | 432 possible noncitizens | Each had voted at some point since 1996 |
| Michigan, 5.7 million ballots, 2024 | 15 appear to have voted | 13 referred for prosecution |
| Utah, 2.07 million registered, 2025 to 2026 | 27 registered | 13 had voted since 2018; 25 more probable |
| Nevada, FBI review of the 2020 rolls | 38 may have voted | Closed in 2026, too late to charge (CBS News's report) |
A false registration becomes a false vote only if someone casts a ballot with it, in person or by mail. Of the more than 1.3 million registration forms ACORN collected in 21 states for the 2008 election, about 400,000 were rejected as duplicates, incomplete or fraudulent, ACORN and its partner Project Vote told the New York Times. In Washington, seven ACORN workers were charged in 2007 over about 1,800 made-up registrations; none led to a vote, the Seattle Times reported.
For fiscal years 2001 through 2017, the Government Accountability Office counted 19 election fraud cases filed by the Justice Department's Public Integrity Section, which supervises election crime enforcement nationally, and 185 filed by the U.S. Attorneys' Offices, about 0.02% of their caseload.
Real Cases
A Kansas law in force from July 2015 let the secretary of state, Kris Kobach, prosecute election crimes himself, a power no other secretary of state held. In 2016 he told the News21 project that double voting was “amazingly frequent”.
In three and a half years he brought 15 cases. Thirteen ended in pleas, twelve by people who had voted in two states and one by a noncitizen; two were dropped. Each plea brought a fine, and nobody went to jail (the Sunflower State Journal’s count). In 2018, at the trial of Kansas’s proof-of-citizenship law, with him as defendant, Judge Julie Robinson found his evidence showed at most 39 noncitizens on the Kansas rolls since 1999, and struck the law down.
Three CasesThe 2020 Litigation Record
Trump and his supporters brought 64 cases in the six battleground states, by the count of Lost, Not Stolen, a 2022 review by eight conservatives, three of them retired federal judges. Fourteen were dropped, 20 were dismissed before any hearing on the merits and 30 were heard on the merits. The one win, in Pennsylvania, involved too few votes to matter.
| Jurisdiction / Case | Primary Allegation | How Decided |
|---|---|---|
| Arizona — Ward v. Jackson | Errors in duplicated ballots | Nine errors in the 1,626 duplicate ballots checked, worth seven votes to Trump and two to Biden. The Arizona Supreme Court affirmed, calling the error “statistically negligible”. |
| Nevada — Law v. Whitmer | Faulty machines, illegal votes | After a hearing, the contestants had proved no machine faults, misconduct or illegal votes “under any standard of proof”. The Nevada Supreme Court affirmed four days later. |
| Michigan — Costantino v. Detroit | Fraud in Detroit’s count | Injunction refused without a trial; the judge found the challengers’ account “not credible”. Both appeals courts denied leave to appeal. |
| Wisconsin — Trump v. Biden | Four groups of absentee ballots | Rejected 4–3. The “indefinitely confined” challenge was “meritless on its face” and the rest came too late; the dissenters wanted the merits decided. |
| Pennsylvania — Trump v. Boockvar | Counties’ differing mail ballot rules | Dismissed with prejudice; the Third Circuit upheld refusing a further amendment as late and futile. |
| Georgia — Wood v. Raffensperger | Absentee ballot and recount rules | Denied for lack of standing and as too late. The Eleventh Circuit affirmed on standing and mootness alone. |
| U.S. Supreme Court — Texas v. Pennsylvania (et al.) | Bar four states from appointing electors | Leave to file denied for lack of standing; Alito and Thomas would have allowed filing but no other relief. |
The 2026 Investigations
On 28 January 2026 FBI agents searched Fulton County’s election hub in Union City, Georgia, and took more than 600 boxes of its 2020 election records, including the ballots, tabulator tapes, ballot images and voter rolls. The case began with a referral from Kurt Olsen, a lawyer who had worked on Trump’s effort to overturn the 2020 result, whom Arizona courts had sanctioned over election suits for Kari Lake, and whom Trump had since made Director of Election Security and Integrity.
The affidavit, sworn by FBI Special Agent Hugh Raymond Evans, listed five “deficiencies or defects” in the count, among them missing ballot images and ballots scanned twice in the recount, both admitted by the county. It named no suspect and said the deficiencies would be crimes if they were intentional. The county called the seizure “dependent on unsubstantiated hypotheticals.”
Fulton’s 2020 ballots were counted three times: in the original machine count, by hand in the audit state law required (the narrow margin made it a full count), and by machine in the recount Trump requested. All three confirmed Biden’s win, and state investigators later found errors, such as batches scanned twice, that did not change the result. On 6 May 2026 Judge J.P. Boulee let the Justice Department keep the records, because the county had not shown that its rights were “callously disregarded,” though he found the affidavit “defective in some respects.” On 7 July Judge William Ray II quashed a Justice Department subpoena for the names and home details of the county’s 2020 election workers, calling its breadth “staggering” and noting that the time limit for any 2020 election charge had long passed. On 19 August Boulee dismissed, without prejudice, a suit to limit use of the seized voter data, Democracy Docket reported. As of September 2026 no one has been charged in the investigation, and the department has made public no evidence of fraud from the records, though it has told a court that some of the seized ballots look suspicious.
The Washington Post reported in February 2026 that the Justice Department had struggled to meet White House demands to prosecute noncitizen voters, slowed, officials said, by a lack of evidence. On 18 September it announced charges against 16 noncitizens in seven states, most of them accused of voting in federal elections. Nevada’s closed review is in the noncitizen table above.
The Limits of the Record and the Strongest Counterargument
A list of proven cases cannot show fraud nobody caught; Heritage says its own database “obviously cannot list fraud that goes undetected.” When three researchers tested the most prominent statistical claims about the 2020 election, from Dominion machines to turnout in Democratic strongholds, they found none “even remotely convincing” (Eggers, Garro and Grimmer, 2021).
Arizona’s Senate paid Cyber Ninjas $150,000 to review Maricopa County’s 2020 ballots and, by the Arizona Republic’s count, spent nearly $1.1m in all by June 2022; Trump-aligned groups gave the firm about $5.7m, by its own account. The firm’s own hand count gave Biden 99 more votes and Trump 261 fewer, a net 360. Wisconsin’s special counsel, Michael Gableman, budgeted at $676,000, cost taxpayers nearly $2.8m by Wisconsin Watch’s count, including a $432,000 records settlement in 2025, found nothing that would overturn the result, and was fired by the Assembly speaker who hired him. Georgia counted its 2020 ballots three times. In Texas, the Houston Chronicle found, the attorney general’s office logged more than 22,000 staff hours on voter fraud in 2020 and resolved 16 cases, all Harris County residents who gave false addresses on registration forms; none was jailed.
Those with the power and the motive to find fraud have found individual cases and, in North Carolina, one scheme that forced a congressional election to be run again. None has found fraud near the scale that would change a presidential result.
The verdict: what the evidence shows
Voter fraud is real, and rare. The Heritage Foundation’s database lists 1,619 cases from 1982 to 2025: about 1,380 criminal convictions, 138 diversion programs, 50 civil penalties and 50 official or court findings. The 22 federal general elections of those years counted about 2.26 billion ballots, so that is about 7 cases for every 10 million, and the cases also come from local, state and primary elections. Heritage says the list is not complete, and fraud that is never caught cannot be counted.
Fraud that was caught has changed a result: an absentee-ballot scheme in North Carolina’s 9th congressional district led the state board to order a new election in 2019. Nothing found comes near a presidential result. Reversing 2020 meant wiping out Biden’s margins in Arizona, Georgia and Wisconsin, 42,918 votes together, and Heritage lists 17 cases from that election in those states. The investigations since, from Arizona’s 2021 review to the 2026 seizure of Fulton County’s records, have produced no evidence of fraud at that scale. The kind voter ID is aimed at, voting in someone else’s name at the polls, is among the rarest: 34 of Heritage’s 1,619 cases.
This article concludes that: (1) documented voter fraud is rare, about 7 cases for every 10 million ballots cast in federal general elections in Heritage’s database; (2) in-person impersonation, the fraud voter ID is aimed at, is among the rarest kinds; (3) no evidence has shown fraud on the scale that would have changed the 2020 presidential result; and (4) fraud can still force a small, close race to be run again, as it did in North Carolina in 2018.
These conclusions would be falsified by:
• A court, audit or prosecution establishing fraudulent votes in the tens of thousands in a single state’s presidential count
• A check of a whole state’s voter file against citizenship records finding illegal votes at many times the rates found so far
• Evidence that undetected fraud is large, such as patterns in audits or recounts that fraud explains and ordinary error does not
