This study compared the ability of a clinical and administrative data base in New York State to predict in-hospital mortality and to assess hospital performance for coronary artery bypass graft surgery. The results indicated that the clinical data base, the Cardiac Surgery Reporting System, is substantially better at predicting case-specific mortality than the administrative data base, the Statewide Planning and Research Cooperative System. Also, correlations between hospital mortality rates that are risk-adjusted using the two systems were only moderately high (0.75 to 0.80). The addition of new risk factors from the Statewide Planning and Research Cooperative System improved the predictive power of both systems but did not diminish the difference in effectiveness of the two systems. The three unique clinical risk factors in the Cardiac Surgery Reporting System (ejection fraction, reoperation, and more than 90% narrowing of the left main trunk) seemed to account for much of the difference in effectiveness of the two systems.