Proxy respondents have often been used in case-control studies of cancer and pesticides. To evaluate the effect of exposure misclassification, we compared data collected during 1981-1983 from participants interviewed for a case-control study of leukemia and non-Hodgkin's lymphoma with data collected during 1990-1991 from proxy respondents for participants who died or became incompetent since the initial interview (328 self-proxy pairs). As questions increased in detail, agreement percentages decreased. Agreement percentages were highest for demographic and general farming information (averages = 88-90%) and lowest for specific pesticide use (averages = 68-74%). Generally, odds ratios calculated from proxy respondent data were less than those from self-respondent data; however, several exceptions occurred. The findings indicate that pesticide data provided by proxy respondents will not necessarily result in the same estimate of risk and/or lead to the same conclusions as data provided by self-respondents.