The model predicting the detonation velocity of explosives was founded on the back propagation (BP) neural-network (BP neural-network has been trained by a hybrid genetic algorithm which based on elitist model algorithm and adaptive crossover mutation), the three-dimension data modeling, molecular weight, oxygen balance and charge density of explosives. The detonation velocity of some explosives were predicted by using the ameliorative BP neural network model. The forecast results indicate that the predicted values by using this model approaches the experimental volues in literature. The absolute errors are ±7%. And there are some analogies between the relative parameters (including the molecular, oxygen balance and charge density of explosives) and the detonation velocity of explosives. The results also show that the yield model has high predicting accuracy. It is a novel method for predicting and estimating the detonation velocity of new explosives.