The accuracy of bedside diagnoses was prospectively studied in 100 consecutive patients admitted to the neurology service at New England Medical Center, Boston. Each patient was evaluated independently by a junior resident, a senior resident, and a staff neurologist, who were required to make an anatomical and etiological diagnosis based solely on the history and physical examination. Fourteen patients were excluded because their diagnoses were known before admission. Of the remaining 86 patients, it was possible to confirm anatomical and etiological diagnoses in 40 by matching the clinical syndromes with highly specific laboratory findings. In the other 46 patients, the diagnoses could not be confirmed because the laboratory studies (including magnetic resonance imaging) were negative or nondiagnostic. In the 40 patients with laboratory confirmed final diagnoses, the clinical diagnoses of the junior residents, senior residents, and staff neurologists were correct in 26 (65%), 30 (75%), and 31 (77%), respectively. There was a trend for error rates to be higher among junior residents than staff (p = 0.06). The errors by the junior residents, [senior residents], (staff) were attributed to incomplete history and examination in 4  (0), inadequate fund of knowledge in 4  (3), and poor diagnostic reasoning in 6  (6). These results indicate that technology is not a panacea for our diagnostic difficulties and that there is room for improvement in our clinical skills, especially diagnostic reasoning.