In this study we investigate the relative performance of six structural credit risk models in predicting bankruptcy. The six models studied (the Merton, Black-Cox, Leland-Toft, Longstaff-Schwartz, flat barrier and Geske models) cover nested assumptions, thereby allowing us to study the factors that affect the accuracy of bankruptcy prediction. Using data from the period 1983-2002 (prior to the Sarbanes-Oxley Act), we compare the performance of these models for various prediction horizons and identify the factors that have the most substantial prediction power of default. Our results suggest that the left-tail probability of equity return distribution plays the most crucial role in predicting default. Moreover, we find that recovery boundary and multi-period default are important factors, while random interest rates are not.