In this paper we consider the enumeration of all solutions of a 2D edge-matching puzzle. We show that a judicious modeling of the problem, combined with the use of appropriate data structures allows obtaining an effective filtering algorithm with complexity O(1) at each node of tree search. Our experiments show the relevancy of the proposed power/complexity trade-off, compared to the results of some of the best available brute force tree search algorithms and to classical Constraint Programming models.