An algorithm for regular and stationary random components of Strapdown INS an-gular velocity measurements error extraction by dithered laser gyros is proposed. The regular component is approximated by one fixed and two exponential time functions with different damping rates (fig. 2). Frequency analysis of the error ran-dom component in actual data of some systems (fig. 1) showed the growth of sig-nal spectral density along with the frequency growth in low-frequency region (fig. 3). It is shown that polynomial filters pass random high-frequency components far less then arithmetical average calculating filters, and provide for reduction of laser gyro drift estimation error (fig.4).