Emotion recognition is an interesting and challenging area under affective computing domain. Emotion plays and interesting role in human life, in fact most of our decisions are also influenced by our emotion at that particular instance in life. This paper explores the use of single modality i.e. using facial expression can significantly contribute to the task of emotion recognition. Convolutional neural network (CNN) an algorithm of deep learning we have used to extract facial features and use these futures for finding out discrete state emotions. Further fine tuning convolutional neural network i.e. by changing parameters and hyper parameters we have experimented facial emotion recognition task in hand on different dataset available publicly. The results obtain during experimentation varies from as low as 45% to 100% accurate, it is important to note that singly modality, number of images, precautions taken during emotion capturing number of classes of emotion, continuous of discrete emotion classification, age & gender of participants play significant role apart from Number of layers, filter size, activation function, and another hyper-parameter of CNN.