To solve the distributed two-stage hybrid flow shop scheduling problem(DTHFSP) with sequence-dependent setup times, an improved shuffled frog leaping algorithm is proposed to minimize the number of tardy jobs and makespan simultaneously. The heuristic method and random method are adopted to initialize the population, population division is done by using population and memory, and then all memeplexes are divided into a best memeplex, a worst memeplex and other memeplexes according to the quality of memeplexes. Different search methods and the number of searches are applied to different kinds of memeplexes and the best memeplex is excluded from the population division. A multi-objective classical algorithm and two algorithms proposed in recent five years are selected as the comparative algorithms, and a variation of the improved shuffled frog leaping algorithm is used for comparison in order to verify the effectiveness of the new search strategy of memeplexes. The comparative analysis based on extensive instances shows that the new search strategy of memeplexes is effective and the proposed algorithm can solve the DTHFSP effectively.