OBJECTIVE To assess the accuracy of plasma LDL cholesterol concentrations estimated by the Friedewald formula and a direct immunoseparation method by comparison with a reference ultracentrifugation procedure in patients with diabetes. RESEARCH DESIGN AND METHODS Fasting plasma samples with triglyceride concentrations < 4.5 mmol/l were collected from 100 patients with diabetes (28 type I and 72 type II) and LDL cholesterol concentrations were compared by the three methods. RESULTS LDL cholesterol values determined by the reference β-quantitation procedure were highly correlated with both the Friedewald formula (r = 0.96) and a direct immunoseparation method (r = 0.92). Calculated (Friedewald) LDL cholesterol coincided with the reference method with < 10% error in 74% of the total diabetic group (82% of type I and 68% of type II diabetic patients). However, agreement between the direct LDL cholesterol and reference methods was significantly less (P = 0.02), with only 44% of patients having an error of < 10% (52% of type I and 41% of type II diabetic patients). The direct immunoseparation method for LDL cholesterol showed a positive bias with increasing triglyceride concentrations, particularly for patients with type II diabetes. CONCLUSIONS In the group of diabetic patients studied with plasma triglyceride concentrations < 4.5 mmol/l, the Friedewald formula provided an accurate estimation of LDL cholesterol. The direct immunoseparation method significantly overestimated LDL cholesterol at triglyceride levels between 2 and 4.5 mmol/l.