The molecular kinetic theory optimization algorithm has the disadvantages of poor optimization accuracy and ease of being stuck into local extremum. A new molecular kinetic theory optimization algorithm with two populations is proposed. In this algorithm, the whole population is divided into two subgroups: the elite subgroup and the ordinary subgroup. The ordinary subgroup uses the search strategy of traditional molecular kinetic theory optimization algorithm to do a wide range search. Through cooperation, the elite subgroup  does a refinement search to improve the convergence accuracy of the algorithm. Information exchange among subgroups is completed through individual migration. Two subgroups complete the search process by cooperation. Test results show that the improved algorithm significantly improves the convergence speed, accuracy and stability.