Abstract—In this paper, extraction of suitable feature vector as well as the analysis and performance comparison of the feature vectors using Hidden Markov Model (HMM) are presented. Extracting suitable features comprising of centroids, hand distance and hand orientations is a necessary step to represent isolated Malaysian Sign Language (MSL) to enable detection of right and left hand blobs. Then, each feature vector is modeled using HMM and trained to produce its gesture class. By increasing the number of states starting from 3 until 57 states, each feature vector is trained using HMM so that in the recognition phase it could give the maximum probability among all the other HMMs for a specific word. The system performance of the recognition step was evaluated for each feature vector from the trained model, starting from separated feature vector, followed by combined feature vectors and finally, the union feature vectors. In the experiments, we have tested our system to recognize 112 MSL and found that the union feature vector gives the best recognition rate, which is 83%.