A stable state of charge (SOC) estimation method which can adapt to variable current environment through adjusting noise covariance is proposed in this study. The accuracy of battery SOC estimation is the important factor in battery management system. First, the cause of instability on SOC estimation using Kalman filter is analyzed, and then an extended Kalman filter (EKF) is used to build up state space equation. The experimental results show that the SOC estimation error within 10%. Meanwhile, the variety of SOC is decreased from 1% to 0.6%