Reproducibility of results from a self-administered questionnaire on dietary habits (the frequency of taking various foods and eating habits), smoking and drinking was examined to study the reliability of the questionnaire, stability of lifestyle, and the validity of the questionnaire in assessing lifestyle as a risk factor in chronic diseases. The study sample included 120 males and 173 females in a rural town in Kyoto Prefecture, who participated in a series of three health examinations in 1988, 1989, and 1992. A survey using the same questionnaire was performed before each health examination. Reproducibility was assessed by correlation (Spearman's rank correlation coefficient or kappa coefficient), exact agreement of category answered, and comparison of mean frequency. An attempt was made to separate reliability and stability from reproducibility using the data from the questionnaire obtained in the three years. Good reproducibility for one-year and four-year intervals was found for foods taken habitually or often (boiled rice, cooked rice gruel with tea, milk, coffee, fruits and bread), and habits (eating breakfast, some eating habits, smoking and drinking). The same was also found for cigarette or alcohol consumption among current consumers except that cigarette consumption over a four-year interval was more likely to change. The reliability of the questionnaire and stability of these items were considered satisfactory, suggesting that the questionnaire was a valid method of assessing long-term lifestyle. Poor reproducibility of results regarding consumption of green-yellow and other vegetables indicated both poor reliability and a change in consumption over four years, and suggested poor validity. However, disagreement in more than one category of vegetables was not large and misclassification in assessing them as risk factors and changes in frequency of consumption may not be large either. Although correlation coefficients for the items of dairy food and eating snacks were fair, disagreement in more than one category was relatively large which may have caused misclassification. The mean frequency of consumption of pickles and fishpaste, ham or sausage significantly decreased over the four year interval. This may reflect the influence of health education on reducing salt intake.