We propose a new hybrid metaheuristic VNSMADS that is an implementation of the Variable Neighbourhood Search (VNS) algorithm with the Mesh Adaptive Direct Search (MADS) as the local search. Because MADS is a Derivative-Free Optimization method, the targeted problems may be non-smooth on black box functions possessing several local optima. Constraints other than bounds on the variables are not considered. VNSMADS is applied on a real-world problem from Finance: the portfolio selection problem using the investor preferences. Numerical results show the efficiency of our approach. Copyright © 2012 John Wiley & Sons, Ltd.