 This paper presents an improved Thevenin model named dual polarization DP model for evaluating lithium ion batteries in electric vehicle applications. The DP model simulates electrochemical and concentration polarization separately using an extra RC circuit. The genetic algorithm is used to identify optimal time constants of the model from experimental data. The dynamic performance and state of charge estimation are evaluated using the dynamic stress test, DST, and robust extended Kalman filter, RECF, approach, respectively. The DP model has the best dynamic performance and provides the most accurate SOC estimation. Sensitivity analysis shows that errors resulting from initial SOC values are significantly reduced and true SOC is convergent within an acceptable error. This article was authored by Jin Qingfan, Han Wenhui, and Rui Shoung.