Surveys regularly measure the frequency with which respondents engage in particular activities, such as attending religious services, exercising, or watching or reading the news. Accuracy depends on the wording and format of the questions. This report assesses responses to two sets of questions on the frequency of religious service attendance that were asked on the ANES 2010 Panel Recontact Survey, which was administered on the Internet. Respondents were randomly assigned either to receive a »traditional« question based on standard GSS and ANES practice, which asked respondents to choose from a list of seven frequency categories, or to receive an "optimized" version of the question that asked respondents to report the actual number of times they attended services during a reference period of their own choosing. The paper will compare rates of item nonresponse, response distributions, elapsed time to complete the questions, and predictive validity for political outcome variables for the two question designs. The paper will recommend the preferred version of the question to use when accuracy is a priority and will assess the appropriateness and cost-benefit tradeoffs of an alternative question format when administration time is prioritized above accuracy.