This paper proposes a new method of gesture spotting based on the Hidden Markov Model (HMM) that extracts meaningful gestures from continuous hand motion. To remove non-gesture patterns from input patterns, we introduce the threshold model that calculates the threshold likelihood of the input pattern and helps to qualify an input pattern as a gesture. The proposed method extracts gestures with 93.38% reliability. (C) 1998 Elsevier Science B.V. All rights reserved.