Algorithm selection is useful in decision situations where among many alternative algorithm instances one has to be chosen. This is often the case in heuristic optimization and is detailed by the well-known no-free-lunch (NFL) theorem. A consequence of the NFL is that a heuristic algorithm may only gain a performance improvement in a subset of the problems. With the present study we aim to identify correlations between observed differences in performance and problem characteristics obtained from statistical analysis of the problem instance and from fitness landscape analysis (FLA). Finally, we evaluate the performance of a recommendation algorithm that uses this information to make an informed choice for a certain algorithm instance.