As part of an effort to assess the most reliable method of obtaining information about long-term dietary intake, a study was conducted of the correspondence between perception of change in diet and measured change in reported diet from 1967 to 1982. Respondents were 1,201 men and women participants in the Tecumseh Food Frequency Study of 1967-1969 who were 45-64 years of age in 1982-1983, the time of the Diet Methodology Study. Perceptions of change were obtained by asking respondents directly if they thought they currently ate each of 12 food groups more often, less often, or as often as they did during the year of their earlier interview. Measured change was obtained by subtracting each respondent's baseline frequency for each of the 12 food groups from his or her current frequency. For 10 of the 12 food groups, correspondence was found between perception of change and measured change; fats and oils added at the table and meat were the exceptions. Significant linear relationships were found between perceptions and measured change for all food groups except fats and oils. However, greatly varying coefficients were associated with terms such as more often or as often as before, thereby limiting the potential utility of combining perceptions of change with current frequencies to obtain a reliable measure of baseline frequencies.