Temperature drift is one of main errors in Fiber Optic Gyro (FOG),which can be divided into two terms:a random error term associated with system error and environmental disturbance,and a trend term caused by temperature change.An improved AR model of FOG temperature drift and its modeling method were proposed.Firstly,AR model of the random error term was established,and then the trend term was introduced in the AR model.The measured FOG temperature drift data was used for modeling and verifying the validity of model.Results show that:1) The proposed model can accurately predict the gyro drift output under different temperature variations,and the predicted value can effectively compensate for the drift error; and 2) The FOG temperature drift after compensation is reduced to lees than 20% of that before compensation.Visual software was developed based on Labview,which is of practical value in assessment and prediction of the gyro temperature characteristics.