A new adaptive nonlinear state predictor (ANSP) is presented for a class ofunknown nonlinear systems with input time-delay. A dynamical identification with neu-ral network (NN) is constructed to obtain NN weights and their derivatives. The futureNN weights are deduced for the nonlinear state predictor design without iterative calcu-lations. The time-delay and unknown nonlinearity are compensated by a feedback controlusing the predicted states. Rigorous stability analysis for the identification, predictor andfeedback control are provided by means of Lyapunov criterion. Simulations and practicalexperiments of a temperature control system are included to verify the effectiveness ofthe proposed scheme.