The paper presents an improved predictive power control for a photovoltaic conversion chain connected to a grid based on finite states space model of the converter. The proposed control algorithm selects the switching state of the inverter that minimizes the error between active and reactive power predictions to their computed values for all different voltage vectors. The optimal voltage vector that minimizes a cost function is then applied to the output of the power converter. Once the proposed predictive strategy is validated, a multilevel converter is then used to improve and highlight the obtained results. The proposed predictive control strategy uses only one sample time prediction and it is very intuitive since it is very simple and provides best performances compared to other modulation techniques. (C) 2013 Elsevier Ltd. All rights reserved.