Without interviewer quality control, interviewer cheating can seriously affect the accuracy of survey results. This paper proposes a method for designing quality control samples which maximizes the probability of detecting cheating for a fixed cost. First, data on interviewer cheating from a recent U.S. Bureau of the Census study are presented. Then a statistical model for describing dishonest interviewer behavior is proposed which assumes cheating is a random event governed by a probability distribution whose parameters depend on the interviewer. These parameters control the frequency and intensity of cheating as well as the geographic clustering of the falsified units. A general quality control sample design and several associated cost models are proposed. A procedure for optimally choosing the sample design parameters according to specific types of interviewer behavior is described. Finally, the procedure is applied to optimize the interviewer quality control system used by the U.S. Bureau of the Census for the Current Population Survey and other current surveys.