A new computationally efficient approach for nonlinear model predictive control (NMPC) presented here uses the factorability of radial basis function (RBF) process models in a traditional model predictive control (MPC) framework. The key to the approach is to formulate the RBF process model that can make nonlinear predictions across a p-step horizon without using future unknown process measurements. The RBF model avoids error propagation from use of model predictions as input in a recursive or iterative manner. The resulting NMPC formulation using the RBF model provides analytic expressions for the gradient and Hessian of the controller's objective function in terms of RBF network parameters. Solution of the NMPC optimization problem is simplified significantly by factorization of the RBF model output into terms containing only known and unknown parts of the process.