An infrared spectrum quantitative analysis model was built based on particle swarm optimization algorithm (PSO)and backward propagation (BP)neural network. This model aimed at three components of gas mixture, with methane, ethane and propane gases included. The concentration of each component ranged from 0.01% to 0.1%. Five features variables were abstracted from 1 866 infrared spectrum data by principal component analysis as the input of the BP network. The gas concentrations acted as the output. PSO was used to optimize the number of neural network hidden layer nodes. Then, the network was trained to construct models for quantitative analysis of these three kinds of gas. The experiment results show that the time taken for optimizing the prediction model by PSO, about 4 600 second, reduced to one fifth of that of ergodic optimizing, which is about 24 500 second. The precision of the model is corresponsive and the structure of the network is approximately the same. So the PSO has definite practical significance and application potential.