Scholars studying political behavior have long been troubled by the fact that survey respondents typically report having voted at a rate higher than citizens in fact turned out on election day. Many observers of this phenomenon have presumed that it reflects intentional misrepresentation by respondents who did not vote and would be embarrassed to admit it. Previous attempts to reduce social desirability bias have not successfully reduced over-reporting of turnout, though. However, no one has tested on a large scale whether reporting turnout anonymously would reduce over-reporting. This is a strong test of whether social desirability plays a role in vote over-reporting, as anonymity has been shown to reduce social desirability bias in other contexts. People are less likely to under-report report socially undesirable attitudes and behaviors (e.g., anger toward affirmative action policies and falsifying tax returns) and less likely to over-report socially desirable attitudes and behaviors (e.g., recycling) when they know their responses are anonymous and cannot be directly linked to them. In order to test this hypothesis more directly, we implemented two techniques that allowed respondents to report anonymously whether or not they voted: the "list" technique and randomized response. We report the results of four studies involving 9 separate national samples of adults (one telephone and eight internet). Our results suggest that the list technique can be successfully implemented in both telephone and internet surveys. However, using the list technique was successful at reducing turnout reports only among respondents interviewed via the telephone, suggesting that social desirability concerns lead people to intentionally distort their direct self-reports in telephone surveys, but not in internet surveys. However, respondents were apparently unable or unwilling to implement the randomized response technique properly in either internet or telephone surveys, suggesting a limit to its utility in surveys conducted in these two modes.