Abstract: Introduces the principle of strap down north finder and studies the question of improving the precision of north finder when under a random distur bance. Neural network has the ability of simulating non linear curves. It can thus simulate the output of the north finder. The paper shows a design for a hybrid filter that combines neural network and low pass filter. When the north finder meets with a random disturbance, the output of neural network will replace the output of the north finder. The result of filtering the real data that has random distrubance shows that this hybrid filter can reduce the influence of random disturbance efficiently.