Abstract: Based on the molecular structural descriptors as follows: intramolecular hydrogen bond, molecular structure, symmetry, oxygen balance, activity index, crowded degree and so on, the step wise regression (SWR) and support vector regression (SVR) approaches were proposed to model the relationship between the descriptors and the impact sensitivity for 156 nitro energetic compounds. The SWR and SVR models were further validated and compared by using 8 independent test samples. The results reveal that lgH50 exhibits a strong negative correlation with the oxygen balance (OB100) and the activity index (F), the more the chemical groups linked with a carbon atom, the greater the impact sensitivity of the explosive; that the trigger bonds of -CH or -OH can cause H50 to decrease; and that the more the intra-molecular hydrogen bonds in unit mass of explosive, the lower the impact sensitivity. The modeling ability of the SVR approach surpasses that of the multiple linear regression (MLR) approach and its prediction accuracy is also superior to that of the MLR approach, which was validated by using the identical training set and test set. The above investigated results demonstrate that the SVR approach is an effective tool to predict the impact sensitivity of explosives and can provide an important theoretical guidance for the design/synthesis of insensitive high energetic explosive.