We compared low-density lipoprotein cholesterol (LDL) values obtained by the Friedewald formula--i.e., total cholesterol minus high-density lipoprotein (HDL) cholesterol minus very-low-density lipoprotein (VLDL) cholesterol (estimated as triglyceride divided by 5)--with those obtained by lipoprotein fractionation, using 4736 specimens. When triglycerides were less than 2.0 g/L, greater than 90% of estimated LDL cholesterol values were acceptable, within +/- 10% of measured values. At triglyceride concentrations of 2.0-4.0 g/L and 4.0-6.0 g/L, only 72% and 39%, respectively, of the estimates were acceptable. LDL values derived from an alternative formula, estimating VLDL as triglycerides divided by 6, were even less accurate. Nevertheless, the use of estimated LDL for risk classification based on the National Cholesterol Education Program Adult Treatment Panel cutpoints of 1.30 and 1.60 g/L was considered acceptable. At triglyceride concentrations less than or equal to 5.0 g/L, 88% of classifications based on estimated LDL (using triglycerides divided by 5) were concordant with those by measured LDL. Eleven percent of classifications were shifted across one cutpoint, evenly distributed between high and low. Fewer than 1% of classifications, all with Type III hyperlipoproteinemia, were misclassified two cutpoints high. Refinements in the estimation model did not substantially improve LDL estimation or concordance of risk classification.