To determine the observer accuracy and interobserver agreement in identifying S4 and S3 by cardiac auscultation and whether they improve with increasing observer experience. Prospective, blinded study. Cardiology and general internal medicine wards in a university-affiliated teaching hospital. Forty patients with a cardiac diagnosis and 6 patients without were studied. Two cardiologists, one general internist, three senior and two junior postgraduate internal medicine trainees, blinded to the patients' characteristics, examined the patients and documented their findings on a questionnaire. Computerized phonocardiogram was obtained in all patients as a gold standard and was interpreted by a blinded, independent cardiologist. The mean positive predictive values for S4 and S3 were 51% (range, 24 to 100%) and 71% (range, 50 to 88%), respectively. The mean negative predictive values for S4 and S3 were 82% (range, 67 to 94%) and 64% (range, 56 to 85%), respectively. The overall interobserver agreements for detecting S4 was K = 0.05 (95% confidence interval [CI], 0.01 to 0.09) and S3 was K = 0.18 (95% CI, 0.13 to 0.24). There was no apparent trend in the accuracy or interobserver agreement with regard to the level of observer experience. The agreement between observers and the phonocardiographic gold standard in the correct identification of S4 and S3 was poor and the lack of agreement did not appear to be a function of the experience of the observers. The overall interobserver agreement for the detection of either S4 or S3 was little better than chance alone.