The classical statement of the recognition principle (after Boardman, Vogt, Milgram and May) that I know is:
Let $X$ be a (group-like) topological space acted on by the little $n$-discs operad, then $X$ is (weakly) homotopy equivalent (as an algebra over the little discs operad) to an $n$-fold loop space.
In particular, the theorem leaves me unsatisfied since there are some spaces which have the homotopy type of an $n$-fold loop space but which are not acted on by the little discs operad.
The "right" statement of the recognition principle should be a statement inside the homotopy category (of spaces and operads), but I have never seen it stated properly.
Does it appear somewhere? Also, is there a modern version of P. May's proof that appears in the geometry of iterated loop spaces?