When the Radial Basis Function Neural Network (RBFNN)is used for the temperature compensation of a laser gyro bias, it shows lower computing efficiency and numerical pathology due to incorrecting selection of an initial center randomly. Therefore, this paper proposes a new RBFNN method based on the Kohonen network and Orthogonal Least Squares (OLS) algrithm. It introduces the principle and modeling steps of the method and designs data collection and temperature compensation experiments of the laser gyro under normal temperature and variable temperature environments. As the method combines the pattern classification capability of the Kohonen network and the optimal choice capacity of the OLS, it avoids the effect of drawback mentioned above, and can quickly and accurately identify the laser gyro bias affected by temperatures. The identification and compensation tests for the laser gyro bias effected by a variety of temperature change situations are performed by the stepwise regression method, RBFNN method and the proposed modified methods in this paper. The test results show that the three methods all have the abilities to identify fairly in the situation of normal temperature;with increasing the rate of temperature change, proposed RBFNN method not only saves time, the compensated laser gyro bias is all also less than 5×10-4(°)/h (1σ), and its accuracy is improved more than 86%. The proposed RBFNN method enhances the stability and effectiveness of identification accuracy, and is suitable for laser gyro bias temperature compensation in a variety of temperature change conditions.