A nondeterministic algorithm that mimics the foraging behavior of ants to solve difficult optimization problems is known as Ant Colony Optimization (ACO). One of the most important problems in ACO is stagnation. Early convergence to a small region of the search space leaves its large sections une xplored. O n t he o ther han d, ve ry slow c onvergence cannot sufficiently concentrate the search in the vicinity of good solutions and therefore render the search inefficiently. Recent s tudies h ave s hown th at s imilarity g rowth in th e population leads to these problems. Twin Removal (TR) has been already i nvestigated to r educe th e s imilarity i n G enetic Algorithm population but not for any of ACO algorithms. In this paper, T R t echnique i s e xtended t o MAX-MIN An t S ystem (MMAS) an d a n ovel an d e ffective TR me thod i s p roposed b y which not only the negative impact of similarity and run-time are reduced, but al so better results t han M MAS without TR ar e obtained i n mos t c ases. Experiments c onducted on TSP benchmarks showed the robustness of the proposed TR method. Results s how that, r emoval of an ts of initial population h aving certain p ercentage of s olution s imilarity would s trengthen MMAS t o perform b etter, accelerating convergence t o best solution.