In almost all tasks performed by a human being implicit handling of objects, specifically grasp objects, is paramount in most tasks and is an area of robotic research. This paper presents the combination of two programming by demonstration techniques to improve the quality of grasp objects. The two techniques are the Gaussian mixture model and task parameterized Gaussian mixture model. One simulated experiment allows us to obtain a database of different grasps to subsequently to apply the proposed algorithm. The obtained results by using a grasp quality metric shows an improvement to the ones generated only using the task parameterized Gaussian mixture model.