In this paper, receding horizon model predictive control (RHMPC) of nonlinear systems subject to input and state constraints is considered. We propose to estimate the terminal region and the terminal cost off-line using support vector machine learning. The proposed approach exploits the freedom in the choices of the terminal region and terminal cost needed for asymptotic stability. The resulting terminal regions are large and, hence provide for large domains of attraction of the RHMPC. The promise of the method is demonstrated with two examples. (c) 2006 Elsevier Ltd. All rights reserved.