Human action recognition is an important area in computer vision and pattern recognition. Its applications include surveillance systems, patient monitoring and human computer interaction. This area was addressed in many studies A lot of approaches have been developed to solve this problem in 2D and 3D spaces. As the area of 3D vision gained a lot of interest lately we focus on this thesis in human action recognition from 3D data. We present a newly-introduced feature of motion history of skeletons in 3D. First, we compute the skeleton for each volume then a motion history for each action. Then alignment is performed using cylindrical co ordinates based Fourier Transform forming feature vector for classification. We investigate the usage of global 3D features. We generate temporal curves of size of the bounding box, volume, surface to volume ratio, and hull compactness and use this curves to describe the actions. Finally we investigated the fusion of these different type features to enhance the classification of actions or better human action recognition system. We evaluate our algorithms on IXMAS and i3DPost 3D datasets. We proved that skeletons produce better results than volumes. And the usage of the global features along with the skeletons or volumes improves the accuracy further more. The result shows that these features improve the recognition accuracy and can be used to recognize human actions independent on view point and scale.