Visual lip reading recognition is an essential stage in many multimedia systems such as “Audio Visual Speech Recognition” [6], “Mobile Phone Visual System for deaf people”, “Sign Language Recognition System”, etc. The use of lip visual features to help audio or hand recognition is appropriate because this information is robust to acoustic noise. In this paper, we describe our work towards developing a robust technique for lip reading classification that extracts the lips in a colour image by using EMPCA feature extraction and k-nearest-neighbor classification. In order to reduce the dimensionality of the feature space the lip motion is characterized by three templates that are modelled based on different mouth shapes: closed template, semi-closed template, and wideopen template. Our goal is to classify each image sequence based on the distribution of the three templates and group the words into different clusters. The words that form the database were grouped into three different clusters as follows: group1(‘I’, ‘high’, ‘lie’, ‘hard’, ‘card’, ‘bye’), group2(‘you, ‘owe’, ‘word’), group3(‘bird’).