Research on the counterintuitive Monty Hall dilemma (MHD) and analogous problems has shown that correct reasoning is rarely observed, even with the help of certain hints. Making the causal structure explicit or presenting probabilities by means of natural frequencies seem to enhance performance, but only to a moderate degree. The present experiments aimed to analyze the usefulness of these hints for solving an analogous MHD in more detail. Results showed that, compared to relative frequencies, natural frequencies improved reasoning, but this effect depended on previous numerical skills. On the other hand, a graph representing the causal structure had no effect, suggesting that numerical representations are more critical for solving the dilemma. Furthermore, success in solving the dilemma strongly correlated with participants' skill in representing probabilities. Hence, an adequate numerical representation seems to be particularly relevant for understanding counterintuitive probabilistic problems.