Summary: A new algorithm for finding minimal perfect hash functions (MPHF) is proposed. The algorithm given three pseudorandom functions \(h_ 0\), \(h_ 1\) and \(h_ 2\), searches for a function \(g\) such that \(F(w)=(h_ 0(w)+g(h_ 1(w))+g(h_ 2(w))) \bmod m\) is a MPHF, where \(m\) is a number of input words. The algorithm involves generation of random bipartite graphs and runs in linear time. The hash function generated is represented by using \(2m+O(1)\) memory words of log \(m\) bits each. The empirical observations show that the algorithm runs very fast in practice.