In existing multi-population cultural algorithms, information are exchanged among sub-populations by individuals, which limits the evolution performance. So a novel multi-population cultural algorithm adopting knowledge migration is proposed. Implicit knowledge extracted from each sub-population reflects the information about dominant search space. By migrating the knowledge among sub-populations at the constant interval, the algorithm realizes more effective interaction with less communication cost. Taken benchmark functions as the examples, simulation results indicate that the algorithm can effectively improve the speed of convergence and overcome premature convergence.