In our previous paper [5], a depth-dependent crossover was proposed for GP. The purpose was to solve the difficulty of the blind application of the normal crossover, i.e., building blocks are broken unexpectedly. In the depth-dependent crossover, the depth selection ratio was varied according to the depth of a node. However, the depth-dependent crossover did not work very effectively as generated programs became larger. To overcome this, we introduce a non-destructive depth-dependent crossover, in which each offspring is kept only if its fitness is better than that of its parent. We compare GP performance with the depth-dependent crossover and that with the non-destructive depth-dependent crossover to show the effectiveness of our approach. Our experimental results clarify that the non-destructive depth-dependent crossover produces smaller programs than the depth-dependent crossover.