In order to solve the problem of complex noise in MEMS attitude sensor, an adaptive algorithm based on Allan variance is designed to estimate the variance of measured noise on-line. The adaptive algorithm based on Allan variance recurrence formula is utilized to track the variance of on-line data. Eulerian angle differential equations are used as the filtering state equations. The main noise parameters of the gyroscope got by Allan variance analysis of gyro static data are regarded as the noise variance of the filtering system, each data is integrated to figure out the attitude angle. Results show that the algorithm can improve angle discrimination accuracy higher than the conventional filtering method, and under the environment of noise fluctuation, the adaptive filtering algorithm is more practical.