During the last years predictive control has received an increasing attention from industry. One of the reasons is that it takes into account the process constraints in a natural way. Nevertheless, there are situations (perturbations, not well defined constraints, etc.) when it is not possible to compute a sequence of future controls such that all the constraints are satisfied, i.e., the problem is not feasible. When this sort of problem appears it is necessary to apply some infeasibility handling procedure that drives the problem to a feasible region. After reviewing briefly some of the different approaches found in the literature, this paper presents a new method for solving the infeasibilities considering a constrained MIMO GPC based controller. The feasibility is recovered applying different techniques or a combination of all of them and the constraints changes are minimised according to a certain criteria. Finally, some computational results are shown.