If we suppose that the central activity of mathematicians is analysis, the search for the conditions of solvability of a problem (and more generally, a search for the conditions of intelligibility of the things that mathematics problems concern), then mathematical reasoning must concern narrative as well as argument. It follows as well that philosophers of mathematics must use historical method as well as logic and the deductive methods of natural science. I illustrate these claims by Andrew Wiles' proof of Fermat's Last Theorem, and criticize Philip Kitcher's ahistorical account of mathematical knowledge.