This investigation addressed three questions about misclassification in a case-control study of risk factors for pancreatic cancer in which all exposure data were obtained from proxy respondents. These questions were: (1) To what degree was misclassification dependent on the type of exposure? (2) To what degree did misclassification vary by the type of proxy? (3) What was the magnitude of the effect of proxy misclassification on odds ratios measured across several levels of exposure? To answer these questions, we interviewed 163 control (index) subjects and next-of-kin (proxy) respondent pairs. Each of the controls and their respective proxies reported the control's use of coffee, cigarettes, and alcohol and weekly exposure to beef, milk, bacon, fruits, and vegetables. Nonspouse proxies misclassified exposures more than spouse proxies with the exception of cigarettes. Cigarette use was the most accurately reported exposure, followed by alcohol, coffee, and foods. For nondifferential misclassification between cases and controls, the slope of a dose-response curve was decreased from 6.6% to 100% depending on the exposure and the type of proxy respondent. Investigators conducting studies using proxy respondents need to recognize that misclassification is a function of multiple factors, including both the type of exposures under study and the type of proxies available.