The CRS operator improves the signal to noise ratio significantly due to the consideration of neighboring midpoints as well as the offset. The determination of the required attributes for the CRS operator is often done by the pragmatic approach to get initial values that are refined by a local optimization. This works reasonable for most parts, however in more complex structures like salt bodies the result is not reliable anymore. Additionally the pragmatic approach does not perform particularly well in the presence of conflicting dips. Therefore we propose to use a genetic algorithm based optimization and show that the stack and especially the determined attributes are significantly better.