Background: The 2018 PhysioNet Challenge utilizes 13 physiological signals collected during polysomnographic sleep studies to classify explicitly-defined arousal regions. The goal is to assign a probability of arousal at each sample for each test subject. Automatic detection of non-apnea arousals may help us better understand various causes of sleep disturbance and advance sleep arousal analysis. Methods: Neural networks possess powerful feature-learning abilities to gain insights from complex datasets, and they have achieved great successful in areas such as computer vision and signal processing. We trained a deep convolutional neural network (CNN) with normalization, pooling, activation and dropout techniques in Python using Keras on top of Tensorflow. We applied all given signals to our CNN in order to account for all potential causes and to attain accurate classification. At this stage, we outputted a probability of arousal for every 500 samples, and then translated this probability to every sample. Results: The CNN was trained on 737 patients’ sleep data and validated on 185 patients’ sleep data. Our preliminary model obtained a testing accuracy of 0.7014 and validation accuracy of 0.7482. However, we expect this number to differ from the official scoring because in our current learning algorithm, we evaluate accuracies based on mean probabilities of every 500 samples instead of the probability at each sample. Discussion: Our preliminary approach shows promise. Over the next few months, we will work on 1) balancing the dataset before feeding it into our CNN, or combining our current approach with other learning algorithms, such as random forest, to more accurately learn from imbalanced dataset; 2) closely examining the effects each signal has on arousal classification and further improve the architecture of our CNN. We will aim to discuss the strengths and limitations of our CNN in sleep arousal classification using a variety of physiological signals.