The work presents a comprehensive methodology for recognition of temporally progressing hand gestures. Motion measurements associated with the hand position, orientation and finger bending are considered as time-series data sets and utilized for the recognition process. In addressing the hand gesture recognition problem in its multi-feature nature, a novel methodology for discovering relevant features for each gesture class is proposed. The two staged comparison approach with the proposed stratification of gesture classes based on their relevant features enabled the methodology to handle the available large number of gesture classes. Gesture comparison is based on a subspace produced by Fisher Linear Discriminant Analysis (FLDA) of temporal features in a manner that rhythmic differences between gesture trials are minimized. Results of the overall methodology have been elaborated for available AUSLAN hand-gesture datasets.