Focusing on small size and uneven distribution of current facial age database, an approach based on a hybrid model combined with classifier and regressor was proposed for facial age estimation. This approach mainly consisted of two aspects: feature learning and estimation method. In the aspect of feature learning, based on an existing Convolutional Neural Network (CNN), an age group classifier and two age estimators were pretrained on the coarse dataset and then fine tuned on the accurate database. In the aspect of estimation method, a coarse-to-fine strategy was adopted. First, a facial images were classified into teenaged, middled-aged, elderly and two overlap groups. Next, the teenaged and elderly groups were estimated by the classifier model, the middled-aged group was estimated by the regressor model, and the two overlap groups were estimated by both models. The proposed approach can achieve a Mean Absolute Error (MAE) of 2.56 on the test set. The experimental results show that the proposed approach can reach a low error under different races and genders.