The U.S. tax system relies on a form of voluntary compliance. Under U.S. law, taxpayers are required to assist tax authorities by reporting their incomes honestly and paying taxes based on their reported incomes. The system is actively enforced by the Internal Revenue Service (IRS) and the courts, who can impose substantial penalties for noncompliance. Nevertheless, tax evasion is a major concern in the United States, and even more so in other countries. 2 Individuals underreport approximately 10.6% of their incomes annually in the United States, and the IRS estimates that tax evasion reduces U.S. tax collections by $190 billion annually (Herman, 1998). Little is understood about the factors that determine how much income individuals report to the government and, of particular interest to policymakers, about the way in which aggregate compliance responds to changes in the economic and policy environment. This paper analyzes a feature of tax enforcement that encourages a particular dynamic tax evasion pattern in the United States. The IRS is careful not to disclose its audit policies, but there is some consensus within the taxpaying public about various aspects of its practices. Tax returns are selected for examination in an initial pass, and if not chosen within a period of about a year, a return is very unlikely to be examined later unless some unusual event draws the IRS's attention to it. U.S. law provides statutes of limitations -three years for minor tax evasions and six years for major tax evasions. The determinants of whether a return will be audited are not known with certainty, but the audit probability is widely believed to be a function of the extent to which a taxpayer underreports income in the current year. Under these circumstances, a taxpayer who evades much of his tax liability in one year has an incentive to report honestly in the following year, since, if he is subsequently audited and caught evading, he is at greater risk of incurring substantial penalties for past evasions.Section 2 of the paper formalizes this notion by introducing a model of tax evasion in which taxpayers who were not audited during the previous year are conscious of their potential liabilities for undiscovered past evasions. As a general matter, these taxpayers are likely to evade less this year if last year's evasion level was particularly high, and conversely, to evade more than usual following a year of little evasion. These general tendencies are tempered somewhat by the dynamic nature of any taxpayer's compliance problem. Higher-than-normal evasion in a year following little 2 The paper uses the term "evasion" to include not only fraudulent evasion but also other forms of willful income underreporting for tax purposes that may differ legally from evasion. See also footnote 10.