In the recent years, the bank in accordance to IRB (Internal Rating Based Approach) of BASEL II, the prediction of probability of default has become the banking circles extremely important topic. Many researches have concentrated on this topic, but the interaction effect of independent variables on dependent variable was rarely considered during the process of significant independent variables selection. This research integrates independent variables with interaction into combined variables, and puts those combined variables into constructing the predictive model of probability of default of personal consumer loans. The result shows that the combined variables can improve the abilities of prediction. This conclusion should be an important reference for banks when construct the predictive model of probability of default.