A device and a method for predicting photograph timing. First, a detection step is performed. Then, a face determination step is performed to determine whether or not a face is shown. A face-classifying step is then performed. If a detected image is a non-front face, a difference generation step is then performed. A difference determination step is performed when a plurality of differences are obtained to determine whether or not the sequentially adjoined differences are reduced, wherein if the sequentially adjoined differences are reduced, a photographing timing is predicted according to a reduction degree of the differences. As a result, the probability of successfully photographing the front face can be enhanced. In addition, a training database that has both front face information and non-front face information can be utilized. As a result, it is not easy to determine the non-front face as the front face incorrectly.