In case-control studies, recalled dietary data from next-of-kin are sometimes used as a surrogate measure of exposure; however, there is limited evidence comparing the ability of study participants and next-of-kin surrogates for the reliability of their responses with respect to past dietary recall. We compared dietary information from 303 subjects who were administered a food frequency questionnaire in 1980 with that from 196 of the same subjects and 107 next-of-kin of deceased subjects 5 years later, but with reference to 1980 diet. Agreement between 1980 and 1985 reporting with respect to food groups, food preparation methods, and adherence to special diets was primarily assessed using the kappa statistic. The concordance between 1980 and 1985 reporting of specific food groups was generally poor. Regarding various methods of cooking meats and the use of different types of cooking fats, next-of-kin respondents showed very poor agreement with the reporting of their deceased relatives, and within-subject agreement was also poor for frying meats, baking meats, and for cooking with margarine and vegetable oil. Subjects and next-of-kin were able to reproduce earlier reporting of a special ulcer diet, but not diabetic or low-salt diets. Overall, subjects tended to have better agreement with their own earlier reporting than did next-of-kin, and spouses were found to be more reliable next-of-kin respondents than other relatives. Dietary data collected retrospectively from next-of-kin may be unreliable.