Age estimation from a facial image is still a challenge due to the variations caused by different aging processes, face appearance, human expression, and face pose. In this paper, we provide a comparative study for age estimation using classic image features as well as deep image features that are provided by pre-trained deep Convolutional Neural Networks. The presented work compares several image features. The experiments are conducted on two face datasets: MORPH II and PAL. In the light of the conducted experiments, image features that are providing the best performances can be highlighted.