Abstract: This paper analyses the intrinsic causes of low efficiency searching the global optimum by genetic algorithms. A class of parallelism evolution technique for niches implemented by crossover of similar individuals and (μ+λ) selected mechanism are proposed. It was proved theoretically and analytically that this kind of niche technique can provide strong selected pressure and also maintain the diversity of individuals in populations. The results of simulation experiments of minimizing discontinuous and multimodal functions with higher dimensions by using genetic algorithms introducing the niche mechanism show that it can remarkably improve the reliability of global convergence and converging velocity, and offer practical facts to justify the design and application of this niche mechanism.