To obtain the prediction model with higher precision and larger suitable scope for predicting impact sensitivity of polynitro compounds, based on the quantitative structure-property relationship (QSPR) principle, genetic algorithm (GA) was adopted in variable selection and GA-multiple linear regression (MLR) model and GA-artificial neural networks(ANN) model were established and employed in research for figuring out the relationship between impact sensitivity and molecular structure of 149 kinds of polynitro compounds. Two model′s correlation coefficient is 0.854 and 0.974, respectively, the root-mean-square error is 0.195 and 0.071, respectively and the mean-absolute-error is 0.157 and 0.051, respectively. By comparing the models, the results show that the prediction precision of two models are higher than the existing QSPR model, and GA-ANN model is obviously superior to GA-MLR model.