When simple genetic algorithm (SGA) is applied to solve structural topology optimization problem checkerboard-like material distribution and disconnection phenomena often occur. To solve these problems, we introduce a stress-based crossover operator (SX) in this paper. Comparisons of SGA, ESO and SX are performed to experiment the effectiveness of SX. Generation alternation models are adopted to speed up the convergence speed. Different elements are discussed to study how the elements affect the resulting topology. At last three mapping methods are used to discuss the stress dependency of SX.