In view of the applications on the small unmanned aircrafts (UAVs), the data noises of Attitude and Heading Reference System (AHRS) has a great influence on the accuracy of attitude estimation of UAVs. This paper analyses the characteristics of the measurement data noises of AHRS in the time domain and frequency domain, and puts forwards corresponding filtering algorithm, combining with attitude estimation algorithm by fusing multi-sensors. Experiments are carried on to verify the effectiveness of the filter algorithm.