Both Convolutional Neural Networks (CNN) and Dictionary Learning (DL) are powerful image representation learning sys-tems based on different mechanisms and principles, so whether we can integrate them to improve the performance is notewor-thy exploring. To address this issue, we propose a novel general-ized end-to-end representation learning architecture, dubbed Convolutional Dictionary Pair Learning Network (CDPL-Net) in this paper, which seamlessly integrates the learning schemes of CNN and dictionary pair learning into a unified framework. Generally, the architecture of CDPL-Net includes two convolu-tional/pooling layers and two dictionary pair learning (DPL) layers in the representation learning module. Besides, it uses two fully-connected layers as the multi-layer perception layer in the nonlinear classification module. In particular, the DPL layer can jointly formulate the discriminative synthesis and analysis representations driven by minimizing the batch based recon-struction error over the flatted feature maps from the convolu-tion/pooling layer. Moreover, DPL layer uses the l1-norm on the analysis dictionary so that sparse representation can be delivered, and the embedding process will also be robust to noise. To speed up the training process of DPL layer, the efficient stochastic gradient descent is used. Extensive simulations on public data-bases show that our CDPL-Net can deliver enhanced perfor-mance over other state-of-the-art methods.