To investigate the relationship between the reproducihility of measurements of exposure from a frequency questionnaire and inferences from a case-control study, 229 patients were reinterviewed on the telephone an average of 6 weeks after the original interview. Several indices of agreement were computed. Information on smoking and use of alcohol and coffee was quite reproducible, as reflected by most measures of agreement, using grouped or continuous data. Frequency of meat, vegetable, and fruit intake had low reproducibility. For all variables, reproducibility was similar for cases and controls. Reproducibility of the exposure variables was a good predictor of the reproducibility of the relative odds (RO) obtained with the first and the repeat interviews. Lack of reliability of the exposure variable appeared to produce fluctuations in the ROs in either direction, even though the same degree of agreement was observed for cases and controls. This situation may partially explain discrepant findings between case-control studies of diet and chronic diseases. (Epidemiology 1990; 1:305–310)