This paper presents recognition methodology of motion from 2 dimensional images. The following two problems are discussed : (1) how to define features for the recognition. (2) how to realize pattern matching. For (1), authors classify features into two : temporal ones and sequential one. Distances are defined with respect to two features. For (2), dynamic programming is introduced because of robustness of phase deflection, expansion and contraction. Recognizing experiments on the finger language showed the effectivenss of the proposed algorithm.