This paper suggests an algorithm to choose good pictures among a bundle of pictures. Since it is very hard to collect private selections of good images from people, the paper divides facial expressions into ten categories, and add ‘good’ tags to photos included in some categories like wink or grin. The proposed algorithm uses convolution neural networks (CNN) to classify the pictures as good or not. The experimental results show that the accuracy of the algorithm is 97.15%, and average execution time is 0.3 seconds. For application purposes, the proposed algorithm is further applied to a group photo in order to count the number of faces that look good.