In 1985, 216 men, participants in the Baltimore Longitudinal Study of Aging, were asked to report on their usual diet during the period 1971-1975. The reference data for past diet consisted of multiple seven-day records collected between 1971 and 1975. The group was randomly assigned to complete the diet history either by personal interview or by mailed self-administered questionnaire. The items and response categories were identical in both cases. For both groups, the mean nutrient intake estimated by questionnaire was within +/- 10 per cent of the past diet record values for most nutrients. When the questionnaire was administered by an interviewer, correlations with past diet were approximately as good as those obtainable with present diet in other studies and indicate that useful studies of past diet can be performed with this questionnaire. Correlations for the mail group were considerably lower and suggest that assessment of past diet using this questionnaire format without personal instruction or callbacks may yield poor results. The effect of age was negligible, but response errors and change in diet since the target period had important effects on correlations with past diet in both interview and mail groups.