The finite-time synchronization problem of a class of nonidentical neural networks with time-varying delay is studied. Firstly, by using the drive-response concept to derive an error system, a suitable integral sliding mode manifold is constructed by applying the synchronization error. If the state trajectories of the error system are driven onto the sliding mode surface, the synchronization error will thereafter converge to zero in finite time. Then, by combining the bounded conditions on neuron activation functions, a proper sliding mode controller is designed. Based on the designed controller and the Lyapunov stability theory, the state trajectories of the error system can be driven onto the sliding mode surface, such that the finite-time synchronization of nonidentical neural networks with time-varying delay can be performed. Finally, numerical simulation results verify the effectiveness of the proposed method.