Medication inventory is more reliable than self-report in assessing prescription drug use in elderly populations. It is not known how strongly medication inventory reflects actual medication use as measured by serum drug levels. In the Cardiovascular Health Study, medication data were collected annually by study interviewers from medication containers brought to the clinic visit. At the fourth clinic visit, venipuncture was performed under 12-hour fasting conditions. Participants were told to take medications as usual. Based on medication inventory results, we randomly selected 55 users and 55 non-users of four cardiovascular drugs: aspirin, propranolol, hydrochlorothiazide, and digoxin. All 110 blood samples for each of the four drugs were analyzed; cut points were based on detectable levels given laboratory limitations. Kappa statistics (K) tested degree of agreement between medication inventory findings and serum detection. Assays were completed on 400 samples (91%). Agreement for aspirin (n=102) was poor: K=0.16 (95% CI: 0.0-0.32). Agreement for propranolol (n = 98) was fair: K=0.43 (95% CI: 0.27-0.59). Agreement for hydrochlorothiazide (n=100) was good: K=0.62 (95% CI: 0.53-0.91). Agreement for digoxin (n=100) was excellent: K=0.94 (95% CI: 0.74-1.0). For four all drugs, lack of agreement was confined primarily to participants who reported use but did not have detectable levels. Excluding aspirin users, only one non-user (0.7%) had drug detected in serum. The medication inventory is a reasonably sensitive and a fairly reliable method for ascertaining non-aspirin cardiovascular drug use in the elderly even though this method may overestimate use as assessed by serum level.