Face emotions are internal feelings of a human which reflects on the human face in terms of specific expressions. It appears naturally rather than through forceful effort and are convoyed by physical variations in facial muscles that implies various expressions on the face. Several standard emotions are happy, anger, sad, fear, surprise, disgust, etc. Facial expressions play an essential role in non-verbal communication. In the area of modeling of intelligent computer vision which can recognize the human’s emotion, various researches have been performed. But advance research which may analyze beyond of ‘just emotion prediction’, for example ’emotion attention’ or ‘salient emotion’ has not been covered much in the literature. In this paper, we are finding the attention scores of same emotion in various levels. Obtained results justifies that, the higher expression level of emotions gives more attention. The experiment has been performed on different levels of same emotions (Frames by Frames) using deep Convolution Neural Network (CNN) and analyzed, how saliency of face is changes with low level to high level of same emotions. FER-2013 and CK+ databases have been applied for training and testing respectively. The proposed approach delivers fairly good result it may give inspiration to the researchers for modeling an intelligent vision system which can predict not only emotion recognition but more than this.