Total variance is used to analyze the random noise of Fiber Optic Gyroscope (FOG), which efficiently solves the problem that the Allan variance calculation is easy to be unstable at long-term τ values. But there is algorithm error with total variance when analyzing the random noise of FOG directly, which cannot really reflect the variances of rate random walk noise, quantization noise and exponentially correlated noise. So an improved method based on total variance is presented to suit random noise analysis of FOG. The estimation performance with the simulated random noise shows that the improved total variance can efficiently identify the noise types and levels. According to the analysis of measured FOG signal, it can improve the confidence in the case of great mean index, the values of variance have a good stability, and it is better and more exact than Allan Variance in identifying noise coefficient.