This paper provides quantitative structure activity relationship (QSAR) models for predicting the radiosensitization effectiveness of nitroimidazole compounds. A new method, combining a heuristic method and projection pursuit regression, was used to build an advanced QSAR model. Compared to the conventional multi-linear regression model, this model showed better predictive ability and reliability, with the values of regression coefficient (R(2)) and root mean square error (RMSE) 0.92 and 0.18 for the training set and 0.90 and 0.17 for the test set, respectively. The provided models were useful tools to predict the radiosensitization effectiveness of nitroimidazole compounds. Also, the new finding descriptors derived from this study will help us to facilitate the design of new radiation sensitizers with better activities.