This paper introduces a region-based approximation method to solve optimal control problems with Approximate Dynamic Programming (ADP). The backbone of the proposed solution is partitioning the domain of training to smaller regions in which the value function varies slowly. Afterward, for each region, a Linear in Parameter Neural Network (LIPNN) is trained to capture the behaviour of the value function in that region. It is shown that the method improves the precision in value function approximation, which leads to improvement in the performance of the closed-loop system. Meanwhile, the possibility of expanding the domain of training in ADP solutions by region-based approximation is discussed. At last, it is shown how the method can potentially eliminate the need for trial & error to select a proper neural network in classical ADP solutions.