Dynamic Allan Variance（DAVAR）is an effective tool for the analysis of non-stationary signal.Since the window function is selected fixed, it has the defects of power leakage and single quantity in the identification of noise.Thus we put forward an improved DAVAR algorithm based on adaptive sliding window taking σ2 ratio (F value) as the criteria.Firstly, the F value is used to reflect the mutation characteristics of stochastic signal.Then, the length of the window function is calculated out according to the real-time change of F value, and stochastic signal is intercepted automatically.Finally, the piecewise Allan value is drawn on 3-D graphics.Analysis is made to the collected gyroscope signal.The result shows that the method can accurately reflect the change of dynamic noise characteristics, and ensure implementation of mutation tracking ability and good degree of confidence, which provides a theoretical reference for the study of inertial navigation system composed by gyroscopes under interference factors.