Past dietary habits are etiologically important to incident disease. Yet the validity of such measurements from the previous 10-20 years is poorly understood. In this study, the authors correlated food frequency results that were obtained in 1994-1995 but pertained to recalled diet in 1974 with the weighted mean of five random 24-hour dietary recalls obtained by telephone in 1974. The subjects studied were 72 Seventh-day Adventists who lived within 30 miles of Loma Linda, California; had participated in a 1974 validation study; were still alive; and were willing to participate again in 1994. A method was developed to allow correction for random error in the reference data when these data had differentially weighted components. The results showed partially corrected correlation coefficients of greater than 0.30 for coffee, whole milk, eggs, chips, beef, fish, chicken, fruit, and legumes. Higher correlations on average were obtained when the food frequencies were scored simply 1-9, reflecting the nine frequency categories. The 95% confidence intervals for 15 of the 28 correlations excluded zero. Incorporation of portion size information was unhelpful. The authors concluded that in this population, data recalled from 20 years ago should be treated with caution but, for a number of important foods, that the degree of validity achieved approached that obtained when assessing current dietary habits.