Aiming at traffic congestion detection, a method of detecting traffic congestion images using Convolutional Neural Network (CNN) is proposed. First, a neural network classification model with three layers of convolutional layers was designed based on the TensorFlow framework. Secondly, the classification model is trained and evaluated using road congestion and non-congestion pictures. Finally, the well trained model is used to carry out road congestion detection. Comparing with the multiple deep learning classification models, the proposed convolutional neural network model shows high efficiency in congestion detection, and the detection accuracy reaches 98.1%.