This paper presents a modified model predictive control (MMPC) for bidirectional power flow control between the electric vehicle (EV) chargers and the main grid. In contrast to the conventional finite control set MPC which selects an optimal switching state from eight possible voltage vectors, the proposed MMPC is based on the application of an optimal voltage vector chosen from an extended set of 20 modulated voltage vectors with a fixed duty ratio. To reduce the computational burden introduced by the increased number of voltage sets, a preselection algorithm is developed for the MMPC method. Six voltage vectors are preselected from the 20 sectors. Due to the increased number of the voltage space vectors, the grid currents and active and reactive power performance can be improved by using the proposed MMPC scheme. Both the conventional and proposed methods are compared through experimental test results of a two-level three-phase off-board EV charger.