In this paper a novel and the new method for face recognition is developed and analyzed using doubly truncated multivariate Gaussian mixture model. The 2D DCT coefficients as the feature vector of the each individual face is modelled by k component mixture of doubly truncated multivariate Gaussian distribution. The number of components and initialization of the model parameter's are obtained by the k-means algorithm and face image histogram. Using the EM algorithm the model parameter's are obtained. A face recognition algorithm is developed by a maximum likelihood function under baysian framework. The efficiency of the developed algorithm is evaluated by obtaining the recognition rates using JNTUK face database and YALE database. This algorithm out perform the face recognition algorithm based on GMM with DCT coefficients.