A major barrier to the conduct and interpretation of retrospective studies of diet and cancer has been uncertainty about the reliability of retrospective measures of diet from the distant past. The authors therefore conducted a study to assess the reliability of retrospective dietary reports and to determine whether the retrospective report or the report of current diet is the better indicator of past diet. Persons (n = 323) originally interviewed regarding their diets in 1975-1979 were retrospectively reinterviewed in 1984-1985. There was little difference between the retrospective reports and the reports of current diet when group means were examined as indicators of past diet. The retrospective reports tended to overestimate the past frequency of consumption for most foods, whereas the reports of current diet tended to yield underestimates. Because food frequency-based dietary history data are more useful for ranking study subjects than for generating estimates of group means, correlation analysis was used as the principal assessment of the reliability of the two indicators of past diet. The retrospective reports more closely correlated with the diet reported at the original interview than did the report of current diet (for 37 of 47 foods). Nutrient indices based on the retrospective history were also more highly correlated with those of the original diet than were indices based on the current diet. No differences were noted in the reliability of retrospective reports according to age or sex. Subjects accurately reported perceptions of changes in their consumption of most foods, yet an estimate of past diet created by adjusting current diet for perceived change did not correlate more highly with the original diet than did the retrospective report. The authors conclude that assessing current diet to make inference about diet from the distant past does not yield more reliable estimates of past diet than does the retrospective dietary history. The best estimate of diet from several years in the past may be derived directly from a retrospective dietary history which focuses on that past period of time.