In order to compensate the high frequency noises and drift errors of fiber optic gyro (FOG) in attitude measurement system under disturbing environment, a new drift error modeling method based on Ⅱ generation wavelet transform and grey neural network algorithm is proposed. The Allan variance method is adopted to analyze the output signal of FOG under disturbing circumstance. Ⅱ generation wavelet transform is applied to separate drift errors and high frequency white noises. After greying drift signal, an Elman neural network for modeling and compensation is established. Experimental results show that, compared to single grey theory model or Elman neural network, the proposed method eliminates white noise effectively, and improves modeling precision up to 96%, which increases the strike precision of combat vehicle.