Pseudo-collision (Pc) as a common but neglected phenomenon in swarm optimization algorithm is revealed in this paper. Mechanism analysis on the inevitability of Pc indicates that both the lack of relation among samples and the unconstrained behavior of sampling are the inherent character of agent operation causing Pc in state-of-the-art swarm algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO). Based on the result of mechanism analysis, a novel partition management and classification sampling strategy is proposed to reduce Pc. In addition, both uniform and non-uniform principles are designed to facilitate the trade-off between exploration and exploitation during optimization. Rain forest algorithm (RFA), of which the evolution mechanism is identical with the above strategy and the principles, is proposed in this paper. By examining the rapidity, accuraty, and generalization capability across six benchmark nonconvex functions, RFA is found to be competitive with or even superior to GA and PSO in dealing with complex multi-peak optimization.