We propose to fine-tune pre-trained deep network models for efficiently classifying fingerprint images. Real datasets of fingerprint images are collected from students at the Can Tho University. We propose to fine-tune recent pre-trained deep learning models such as VGG16, VGG19, ResNet50, Inception-v3, Xception for classifying fingerprint images. Empirical test results on real datasets of fingerprint images show that the fine-tuning strategy for pre-trained deep learning models outperforms the classical approach, i.e. Support Vector Machine (SVM) models trained on Scale-Invariant Feature Transform (SIFT) and Bag-of-Words (BoW). Fine-tuning many last layers (FTm) of deep networks also improves the classification correctness compared to fine-tuning only the last layer in networks. FTm-Inception-v3, FTm-ResNet50 give accuracy of 99.25% and 99.21% for the fingerprint image dataset with 559 classes.