This paper presents a method of improving the efficiency of DisCSP algorithms. It is based on a new approach in nogood discovering and learning. The proposed technique is based on hyper-resolution and it involves a specific way of generating more powerful nogoods and reducing their number. Moreover, the possibility of storing and re-using nogoods from one instance of a problem to another is taken into account. A brief presentation of the used simulation tool is included and the method is tested on a manufacturing scheduling problem.