The authors studied temporal change in the reproducibility of a self-administered food frequency questionnaire. During 1988-1994, 492 residents of a rural Japanese town completed five questionnaires including 27 food items, with intervals ranging from 2 weeks through 5.5 years. The reproducibility decreased steadily over time for pairs of the questionnaires administered in the same season (median Spearman's r at 2 weeks and 5 years = 0.62 and 0.28, respectively). The reduction was less marked for those surveyed in different seasons (median r at 5 months and 5.5 years = 0.35 and 0.28, respectively). The short-term, different season correlation at 5 months was lower than the short-term, same season correlation at 1 year. For individual food items, a lower initial reproducibility, infrequent consumption, and a larger difference in seasonal intake were associated with a greater reduction in reproducibility over time. The results indicate that reproducibility studies should deliberately choose the intervals and the seasons for surveys.