Abstract: Kalman filter is the most usual optimization filter with some limitation when applied in the integrated navigation system. Especially in simple global positioning system/ dead reckoning system (GPS/DRS ),the receiver and the inertial navigation units are low-cost and low-accuracy. To improve accuracy of the system, it must be focused on the advanced algorithm. To compensate the divergence of Kalman filter, a hybrid algorithm composed of back propagation (BP) neural net and genetic algorithm (GA) together with Kalman filter was applied in low-cost GPS/DRS integrated navigation system. This algorithm owns not only self-study ability and good real-time performance of neural net but also optimization assessment ability of Kalman filter, and even overcomes many flaws of neural net,such as slow convergence ,sensitivity about the study parameters and local extremums. The simulation result also proves that this algorithm is prior in precise and stability compared to usual Kalman filter, for example, the statistic analysis shows that the maximal error of latitude is reduced to a lower magnitude.