BACKGROUND Discontinuation rates for drugs used to treat chronic conditions may affect the success of therapy. However, the discontinuation rates reported in clinical trials may not reflect those in primary care settings. METHODS We conducted a cohort study using computerized research files and medical records on 2369 new users of antihyperlipidemic therapy at two health maintenance organizations (HMOs) from 1988 through 1990. The rates of drug discontinuation in these primary care settings were compared with the rates reported in clinical trials published from 1975 through 1993, located with the Medline data base. RESULTS In the HMOs, the one-year probability of drug discontinuation was 41 percent for bile acid sequestrants (95 percent confidence interval, 38 to 44 percent), 46 percent for niacin (95 percent confidence interval, 42 to 51 percent), 15 percent for lovastatin (95 percent confidence interval, 11 to 19 percent), and 37 percent for gemfibrozil (95 percent confidence interval, 31 to 43 percent). For the bile acid sequestrants, niacin, and gemfibrozil, the risks of discontinuation were substantially higher in the HMOs than in randomized clinical trials, in which the summary estimates of this risk were 31 percent, 4 percent, and 15 percent, respectively, for trials of one year or longer. The rates of discontinuation in open-label studies were similar to those in the HMOs. CONCLUSIONS The discontinuation rates reported in randomized clinical trials may not reflect the rates actually observed in primary care settings. The effectiveness and tolerability of antihyperlipidemic medications should be studied further in populations that typically use the agents.