There is good reason to believe that, if it can be decided at all, the realism debate must be decided on a posteriori grounds. But at least prima facie the prospects for an a posteriori resolution of the debate seem bleak, given that realists and antirealists disagree over two of the most fundamental questions pertaining to any kind of empirical research, to wit, what the range of accessible evidence is and what the methodological status of explanatory considerations is. The present paper aims to show that, while the difficulties that face an empirical approach to the realism debate are not to be discounted, they are not insurmountable either. Specifically, it presents a broadly Bayesian strategy for resolving the debate that is capable of solving those difficulties. The strategy crucially involves answers to the aforementioned questions that diverge from both the standard realist and the standard antirealist answers, but that should appear more natural and plausible than those to realists and antirealists alike.