Abstarct: One of the important ways of human’s communication is established by face and its exxpressions. Human exxpression plays an important role in transmission of inner emotions and improvement of quality of the human’s communications. Nowadays, in industrialized world, automatic recognition of face exxpression covers a broad spectrum ranging from psychological and legal studies, animation synthesis, robotic, images comprehension and video conference, communication and broadcasting networks, recognition of suspicious treat for taking security and antiterrorist measures, to human-machine interface. From thirty years ago till now, Scientifics have done many researches and activities in this field and they have achieved salient successes in advancement and improvements of these systems. As we can see they could achieve the recognition rate up to 80 percent in recognition of seven exxpressions of human both in real time-dynamic image and static and record images. All of these systems are baxsed upon three sections as features selection, features extraction, and features classification. From one point of view, facial feature extraction methods are categorized in holistic and analytic methods. From other point of view, these methods are categorized in feature-baxsed, model-baxsed, and image baxsed methods. In this thesis after deep studies, it founds that a lot of current techniques either have mathematical complexity or low recognition rate. So, finding a tradeoff between these two is the biggest question, therefore by following current algorithms, we proposed several supplementary proposal to optimize former methods and finally we approached to an algorithm with a recognition rate up to 99 percent accuracy in sadness or happiness exxpression recognition. At first step, we used modified Loupias salient point method for separating the basic element of face. Feature extraction was done with using principal component analysis and Gabor filters and principal component analysis. Distance criterion and artificial neural network and statistical network were used for feature classification. Recognition rate of 99% was achieved of using Gabor filters and principal component analysis on eyes and mouth that were achieved of Loupias salient point method, and perceptron neural network for classification. Proposed statistical network method announce broad field for more scrutiny in future researches.