Accurate estimations of cell state-of-charge (SoC) for multi-cell series-connected battery pack are remaining challenge due to the inconsistency characteristic inhabited in battery pack and the uncertain operating conditions in electric vehicles. This paper tries to add three contributions. (1) A data-driven filtering process is proposed to select one represented cell to typify the voltage behavior of battery pack. (2) An improved battery model considering model and parameter uncertainties is developed. (3) An adaptive SoC estimator has been developed, in which the SoC of each cell in battery pack can be accurately predicted. The SoC of battery pack can be located with the SoC values of each cell. It significantly improves the safety operation of battery. The result indicates that the estimation errors of voltage and SoC for all the LiPB cells are less than 3% even if given big erroneous initial state of estimator.