When switched reluctance motor was in the status of slow running under direct torque control, calculation of flux was greatly influenced by resistance. In order to solve the issue above. The study observed and analyzed carefully about the relation between resistance and phase current, through comparing the output current between resistance variable motor model and actual motor model, proposed a solution of resistance estimation based on optimized BP neural networks. Optimized BP neural networks had sufficient mathematical theory, with simple structure and clear algorithm. The algorithm based on BP neural networks could recognize variable stator resistance. Put this algorithm into action in the Simulink control system, then comparing the test results between with resistance estimation and without resistance estimation. Experimental results showed that this resistance estimation method could improve system performance when the switched reluctance motor was in the status of slow running.