Abstract: Poor selectivity and cross sensitivity are problems of gas sensors,general BP(back propagation) neural network has low recognition ability and its error is too high to accept.A hybrid two layers neural networks is developed to improve the quantitative analysis of general BP neural network.Using two layers neural networks,three mixed inflammable gases in mine are adopted for the quantitative analysis.The result shows that recognition ability of the neural network is increased and the maximum relative error of quantitative analysis is 4.4%,the precision of quantitative analysis is increased.