Remaining useful life (RUL) prediction of lithium-ion batteries is a key parameter in battery management systems. The accurate prediction of RUL is a prerequisite to ensure the safety and reliability of the battery system. However, it is difficult to use in practical applications, address to this problem, nonlinear autoregressive models with exogenous inputs (NARX) dynamic neural network is proposed to predict RUL. Firstly, according to the discharge data of the lithium-ion battery, three indirect health indicators of cut-off time, constant current time and peak temperature time in discharge are proposed and grey relation analysis (GRA) is used. Then, based on NARX dynamic neural network, the lithium-ion battery RUL prediction model is established. Finally, through comparative analysis with the classic?back propagation neural network based on particle swarm optimization (BPNN-PSO), least square support vector machine (LS-SVM), Closed-loop NARX and Open-loop NARX, It is verified that the proposed method has superiority in the RUL of lithium-ion batteries.