An efficient hybrid method to optimize the phase states distribution, or phase diagram, of a digitally-reconfigurable reflective cell is presented. It allows minimizing phase quantization errors in applications such as reflect arrays. Digital control of the reflected wave phase is achieved by placing a given number of control elements within the reflecting cell (e.g. PIN diodes or MEMS). The method is based on a perturbation of the control elements positions and combines a genetic algorithm and a least squares optimization. It solely requires running a very small number of full-wave simulations prior to the optimization process. The method is illustrated by the optimization of a reflective cell based on PIN diodes technology, but could equally be applied to a MEMS-based digital implementation.