A quantitative history of current dietary intake based on 83 food items was administered by interview to a representative sample of 4,809 subjects in Hawaii in 1977-1979. In 1983-1987, this history was readministered to the 131 original respondents who had subsequently developed cancer, as well as 413 randomly selected subjects who remained cancer free. A surrogate was interviewed when the original subject was unavailable. The repeat interview elicited information about diet at the time of the original interview. The authors found that recall values for macronutrients were consistently higher than original levels for both cases and noncases, which may be due in part to a modification in the administration of the repeat questionnaire. Although there were no marked differences overall between cases and noncases in the ability to recall past diet, differences between the two groups were seen in certain subsets of the sample. In the subgroup with the longest recall interval (8-10 years), cases were not able to recall their diets as well as noncases. Also, the difference between original and recall values was larger for cases with colorectal cancer and all cases diagnosed with distant stage disease, compared with noncases. This was not true for cases of breast and prostate cancer and those with localized or regional disease. The following variables were found to have no significant effect on recall for cases or noncases: sex, age, education, and type of respondent (surrogate or subject). Of the five major ethnic groups included in the study, Japanese had the best recall of their past diets, while respondents reporting a diet change between the interviews had poorer recall than did those who reported no change. These results suggest that differential misclassification in dietary case-control studies may pose a significant problem in certain instances, but that these studies can yield meaningful results with certain constraints on the study population.