We present a real time heuristic learning algorithm. This algorithm is characterized by the complete heuristic learning process, which consists of state selection, heuristic learning, and search path review. The execution of the search path review is controlled by the user specified heuristic learning threshold. The algorithm will return an optimal solution with zero threshold, and near-optimal solutions with non- zero thresholds. We base on the dynamic nature of the resources of a project scheduling problem to present an application approach, which includes definition of states, state transition operator, and the cost of transition between states. Important results with the Patterson’s110 problems are presented.