Aiming at the poor performance of chi-square fault detection for soft fault detection, even leading to the filter divergence problem, a joint fault detection method based on D-S evidential reasoning is proposed. Each sub filter in integrated navigation is used as the evidence. The state and the covariance of each sub filter are used for constructing the fault detection function and calculating the basic belief assignment. Then the fault detection is determined based on D-S fusion results. Simulation results show that the hard fault detection performance of this method has similar quality with the chi- square method. The soft fault detection performance is superior to the chi-square method, which can improve the reliability and accuracy of the integrated navigation system.