Abstract: In view of problems of big error and complex calculation with the least square method and problems of slow network learning speed and being easy to fall into local minimum with traditional BP neural network method when fitting output characteristic curve of infrared sensor, the paper analyzed fitting effect of improved the least square method and improved BP neural network method based on genetic optimization, indicated the improved BP neural network has higher fitting degree and gave experiment result of the improved BP neural network method in detection of methane volume fraction. The result showed that the method can fit ideal curve which improves detection precision and respond speed of infrared sensor.