To improve the accuracy and sensitivity of methane sensor,the improved radical basic function neural network (RBFNN) algorithm is applied to the methane sensor system to correct the nonlinearity.The effect of temperature on the prediction of methane concentration is analyzed,and then RBFNN is used to do the discrete training.The experimental results show that the obtained data is closer to the true values than the actual measured concentrations through RBFNN,which causes that the average error is less than or equal to士0.1%.The effect on the prediction is so good that it achieves the desired specifications, which improved the accuracy and sensitivity of methane detection greatly.