Hybrid genetic algorithms, which are based on steepest descent algorithm and genetic algorithm, are investigated for the purpose of multimodal optimization. The performances of the hybrid genetic algorithms are evaluated with criteria such as convergence probability, average convergence time and average convergence value of the function in the case of solving global optimization for Schaffer function. It is shown that the performances of the hybrid genetic algorithms are better than steepest decent algorithm or genetic algorithm, and the hybrid genetic algorithm, in which the individuals used for local optimization by steepest decent method are chosen by chance in each generation population, is more efficient than that in which the individuals used for local optimization by steepest descent method are selected from excellent individuals.