The main objective of this project is to disentangle two sources of risk premia in CDS spreads: default risk and loss risk. One key challenge in previous studies is lack of a good model to separately identify these two risks from CDS spreads. In this project, we develop a novel CDS pricing model, which inherently incorporate stochastic loss expectation. Numerical experiments show our model is able to reliably estimate the loss expectation as well as default risk from the CDS spreads data. We apply the model to over 600 firms' CDS data to draw a comprehensive picture of the loss expectation across different sectors and over at least 6 years' period.