Inconsistent results among case-control studies of diet and cancer may reflect discrepancies between dietary reports based on current habits and actual intakes from the time of cancer initiation in the distant past. To examine the validity of reported past and current diet as measures of past diet, the authors queuried 873 men and women in three northern California communities about their diet 11 years earlier in 1972 and about their current diet in 1983. They used the same dietary history questionnaire that had been administered to the subjects in 1972. Retrospectively recalled and currently reported nutrient intakes and measures of body size were compared with original reports obtained in 1972. Mean values of recalled nutrient intakes agreed with mean values of intakes reported in 1972. The agreement held for each sex, community, and nutrient examined. By contrast, mean values of current intakes were smaller than those reported In 1972. Correlation coefficients showed closer agreement between recalled and original intakes than between current and original intakes. Nevertheless, recalled intakes correlated more closely with current intakes than with original ones. Analysis of variance was used to partition the variation in recalled and original intakes into components due to interpersonal variation in true intakes, errors in recall, and residual reporting error. Interpersonal variation accounted for only 20–40% of the variation in dietary intakes, with most of the balance due to reporting error. For comparison, interpersonal variation accounted for 70–85% of the variance of recalled and original reports of body size. These results suggest that inconsistencies among case-control studies of diet may be due to large random measurement errors in individual dietary intakes. The apparent validity and reproducibility of mean dietary measures, averaged over large populations, suggests the need for new designs for studies of diet and disease.