Associations between plasma lipoprotein subfractions and changes in coronary artery diseases (CAD) were examined in 74 men who completed the St. Thomas' Atherosclerosis Regression Study (STARS). Plasma lipoproteins were isolated by stepwise, preparative ultracentrifugation at repeated intervals during the 38-month trial. Paired coronary angiograms were quantitatively analyzed by a computerized method. In univariate linear regression analysis, changes in mean absolute width (delta MAWS) and minimum absolute with (delta MinAWS) of coronary segments were significantly correlated with in-trial concentrations of cholesterol in intermediate-density lipoprotein ([IDL] d = 1.006 to 1.019 kg/L), low-density lipoprotein ([LDL2] d = 1.019 to 1.040 kg/L; LDL3, d = 1.040 to 1.063 kg/L), and high-density lipoprotein ([HDL3] d = 1.125 to 1.210 kg/L) subfractions; no significant associations were found with other lipoproteins. IDL, LDL3, and HDL3 cholesterol were then selected for multiple linear regression analysis because these variables were not co-correlated and because they attained a significance of P less than or equal to .1 in univariate regression. In this analysis, only LDL3 cholesterol level was a significant negative predictor (P < .05) of both delta MAWS and delta MinAWS; a positive association between delta MinAWS and HDL3 cholesterol level just failed to reach conventional statistical significance (P = .066). Correlations between changes in coronary luminal dimensions and LDL3 cholesterol level were independent of age, smoking, weight, and blood pressure. Most patients showing regression of coronary atherosclerosis had an LDL3 cholesterol level of less than 1.8 mmol/L. The findings suggest that LDL3 is the plasma lipoprotein subfraction that exerts the single most powerful effect on the course of CAD in middle-aged men with hypercholesterolemia.