A performance test bench for power batteries in hybrid electric vehicles is built. According to vehicles’ real driving condition, the correlative charge and discharge experiments of power batteries are carried out in different environmental temperatures. A new algorithm about self-correction fuzzy neural network control is used to calculate the state of charge(SOC)of power batteries at 25 ℃, and the result is compared with that of Arbin test instrumentation. Theoretical analysis and experimental results suggest that the accuracy of SOC calculated by the algorithm of correction fuzzy neural network control under variable rate discharge for power batteries meets the requirements.