This paper investigates the determinants of the cross-sectional variation in the credit spread regression performance. We provide new evidence that risk premium component is negatively associated with the performance in the firm level. This is somewhat contradictory to the finding of existing studies showing that, in the aggregate level, the risk premium is not crucial in solving credit spread puzzle. Our empirical results show that structural variables inspired by theory are more likely to fail in accounting for the CDS spreads for firms with a higher risk premium fraction. While liquidity and quality are also important factors, the proportion of risk premium is significant even after controlling for the factors.