In this paper, we propose a genetic algorithm, called GenSPN, for finding highly probable differential characteristics of substitution permutation networks (SPNs). A special fitness function and a heuristic mutation operator have been used to improve the overall performance of the algorithm. We report our results of applying GenSPN for finding highly probable differential characteristics of Serpent block cipher. A comparison of the resultant characteristics with the previously published works shows that GenSPN can find differential characteristics of higher probabilities.