Abstract: The temperature characteristics of SOS pressure sensor was analyzed and found that the sensor output is linear and easy to be affected by ambient temperature over a wide measuring range. Based on the conjugate gradient algorithm for neural network, a temperature compensation method for sensor is put forward the way the approximability of the algorithm to any nonlinear function is utilized to drill the neural network so as to enable it to be set up at different ambient temperatures, thus allowing the sensor output can be in a nonlinear mapping relation to the voltage values the sensor actually sensed. The simulation results on computer showed that the method can not only eliminate the influence of ambient temperature fluctuation but obtain the expected linear output from the output terminal of neural network.