We propose a simple and sufficiently fast separation procedure to identify cover inequalities for the multidimensional knapsack problem. It is based on the solution of a conventional integer programming model. Solving this kind of integer programs are usually considered expensive and the proposed method may have been overlooked because of this assumption. The results of our experiments with a small set of randomly generated problems and problems taken from the literature indicate that the method may be a reasonable alternative to the one currently in use.
Forthcoming in Computers and Operations Research (http://dx.doi.org/10.1016/j.cor.2005.05.032).