A cohort of subjects born 1908 to 1925 answered a questionnaire on dietary habits in 1967 (original information). Of the cases of colorectal cancer that were diagnosed in this cohort 1968-1983, 50 were still alive in 1987. A total of 150 controls were selected as a stratified (age, sex) random sample of those who did not have a colorectal cancer. In 1987, the cases and controls were asked to fill in a questionnaire on dietary habits in 1967 (retrospective information) as well as on their current dietary habits. The questions were identical and used to compare retrospective and original information on diet among cases and controls, taking into account any changes in dietary habits during the intervening period. Cases and controls usually had a similar tendency to overestimate or underestimate their previous food intake. For both cases and controls, the tendency to overestimate previous intake was closely related to an increase in consumption during the intervening period and vice versa. Among those subjects who had not altered their consumption during the intervening period, there was usually a rather good agreement between the retrospective and original information. This was true for the cases as well as for the controls. For certain food items, the results suggest that the use of retrospective information would introduce a differential misclassification. In order to avoid or reduce a differential misclassification introduced by differences between cases and controls in changes in consumption due to the disease under study, it may be advisable to interview the cases as soon as possible after diagnosis or even at the time of preliminary diagnosis.