Change and correlates of change in high and low density lipoprotein cholesterol (HDL and LDL) were studied in a prospective population-based study of 614 men and women in Rancho Bernardo, CA, 1972-1980. Correlations between baseline and follow-up examination an average of six years later were 0.72 for HDL and 0.68 for LDL, and 86% of HDL and 89% of LDL values were within +/- 30% of their original values. Predictions of HDL and LDL change were determined by multivariate analysis. Regression to the mean was the strongest single predictor of lipoprotein change and was more important than behavioral change. Among the behavioral variables, HDL change was positively associated with change in alcohol use in both sexes and with change in postmenopausal estrogen use in women and was inversely associated with change in obesity index in men. LDL change correlated positively with change in obesity index in both sexes. Change in reported cigarette use or exercise was unrelated to changes in HDL and LDL in this analysis. The similarity of lipoprotein values at baseline and follow-up suggests reasonable reliability and prognostic validity for a single HDL or LDL measurement. The multivariate analysis results generally confirm current cross-sectionally derived concepts about behavioral correlates of lipoproteins. However, the degree of regression to the mean indicates the wisdom of repeat measurements, particularly for persons with extreme values.