In order to improve the precision of fiber optic gyroscope (FOG), multiple linear regression has been widely applied into identification and compensation of FOG bias drift causing by temperature variation. There exist inconsistencies between the measured temperature and the actual temperature of the FOG’s sensing unit. In this paper, the time delay between FOG bias data and temperature signal is considered before modeling the multiple linear regression for FOG. Then, by using FOG data collected at different time, we can establish more than one regression model. After that, these regression parameters are smoothed to obtain a comprehensive model, which has good parametric adaptability. Experimental results, on periodic and nonperiodic temperature cycling, show the superior performance of this proposed method in compensation of FOG temperature drift and improvement of FOG accuracy. The bias stability of FOG is superior to 0.01°/h at full temperature range -40°C ~ +60°C with temperature change rate -1°C/min ~ +1°C/min.