Background: Randomized clinical trials may not accurately predict drug benefit in clinical practice because the patients and conditions of therapy almost always differ between those settings. Objective: To determine whether differences in low-density lipoprotein cholesterol (LDL-C) time curves between clinical trial results and usual care experience after initiation of statin therapy could be explained fully in terms of observable characteristics of patients and practice in usual care. Methods: We compared LDL-C measurements for 3 years after initiation of statin treatment in individuals enrolled in a health maintenance organization (HMO), with enrollees in the active treatment arm of the CARE (Cholesterol and Recurrent Events) trial. Analysis of the determinants of variation in LDL-C in the HMO cohort was used to adjust the crude results to the distribution of patient and treatment characteristics in the trial. Results: The mean percent decrease in LDL-C was lower in the HMO cohort (n = 1245) than in the clinical trial at the end of each 6-month period, with the difference diminishing over time. Adjustment of the HMO time curve to the baseline characteristics of the clinical trial cohort did not significantly change the mean estimates at any time point. Assuming optimal adherence in the HMO cohort raised the curve over time, with 95% confidence intervals including the means observed in CARE after 18 months. Fixing treatment to pravastatin 40 mg/day, as in CARE, brought the estimates in the HMO cohort very close to those of the clinical trial, with 95% confidence intervals including the means observed in CARE at all time points. Conclusions: Treatment selection, dosing, and adherence under usual care were the primary reasons for which improvements in LDL-C in practice fell short of expectations that are based on clinical trial findings for statin therapy. Beginning with a low dose of statins and titrating to a satisfactory response can delay effective treatment by 18 months or more. Poor adherence accounts for a further substantial shortfall from maximal effect. Differences between trial populations and the general population of statin users with respect to age, gender, and baseline LDL-C have no measurable impact on discrepancies between predicted and observed LDL-C improvement in usual practice.