A comparison of two computer programs for ECG interpretation was undertaken. Twelve-lead ECGs from 300 patients with various clinical abnormalities were interpreted at the National Institutes of Health using version 1 of the IBM program and corresponding orthogonal three-lead ECGs were analyzed by the Glasgow Royal Infirmary (GRI) program. Interpretations were compared with respect to the clinical documentation, wherever possible, and with each other directly in the case of diagnostic statements for which non-ECG documentation was not available. The two programs had a similar performance in determining abnormalities such as myocardial infarction and ventricular hypertrophy. However, with respect to conduction defects and ST-T-wave statements, certain discrepancies between the two program performances were revealed. There were 222 disagreements between various diagnostic statements. GRI was judged correct in 119 of these disagreements and IBM in 70. In these 189 cases the disagreement could most often be accounted for by different criteria and/or algorithms in the two programs or by the use of different ECG lead sets. The remaining 33 disagreements had to be classified as inconclusive.