How large can a subset of the first N positive integers be if the set contains no triple of points in arithmetic progression? We discuss Behrend's construction of a fairly large set with no such triple. We slightly reinterpret Behrend's ideas, and thereby extend the range of applicability of his theorem.