A new image dataset is introduced for evaluating cervical disease classification. The pyramid feature descriptor encodes complementary hand-crafted features. The deep convolutional neural network automatically learns features from images. Results using seven classic classifiers serve as the baseline for future research.