Accurate residual capacity (RC) estimation of batteries is crucial for safe operation of electric vehicles. However, current estimation method has poor generalization ability and low accuracy, which restricts the promotion and development of electric vehicles. This paper puts forward a data-driven method for accurate RC estimation of lithium batteries. Firstly, 103 LiNCM batteries are tested and analyzed under different temperatures and charging rates. Then, incremental capacity curve is obtained and smoothed thorough Kalman filtering algorithm. The aging features obtained from partial charging curves based on incremental capacity analysis is taken as the input of genetic algorithm-back propagation neural network (GA-BP) model. And the corresponding capacity is taken as the output to train estimation model. The accuracy of RC estimation under different working conditions is verified, and the effectiveness of the estimation method in LiFePO4 battery is also verified. The maximum error of the proposed estimation method is less than 2%, and the average absolute error and root mean square error are both less than 1.5%.