The Gerchberg-Saxton (GS) iteration algorithm is widely used for phase-only computer-generated holograms. However, as the GS algorithm was originally developed for single-plane holographic reconstruction, its reconstruction capability for multiple planes is limited. The additive cross-talk between different image planes results in image-quality degradation. In this study, we propose a dynamic compensatory GS algorithm, where the weighting factors of the amplitude-constraining function are dynamically adjusted during each iteration. Both simulation and experimental results prove significant improvement on average image quality of multiple planes.