A heuristic is given for finding minimal perfect hash functions without extensive searching. The procedure is to construct a set of graph (or hypergraph) models for the dictionary, then choose one of the models for use in constructing the minimal perfect hashing function. The construction of this function relies on a backtracking algorithm for numbering the vertices of the graph. Careful selection of the graph model limits the time spent searching. Good results have been obtained for dictionaries of up to 181 words. Using the same techniques, non-minimal perfect hash functions have been found for sets of up to 667 words.