We design and implement a temporal convolutional network model to predict sepsis onset. Our model is trained on data extracted from MIMIC III database, based on a retrospective analysis of patients admitted to intensive care unit who did not fall under the definition of sepsis at the time of admission. Benchmarked with several machine learning models, our model is superior on this binary classification task, demonstrates the prediction power of convolutional networks for temporal patterns, also shows the significant impact of having longer look back time on sepsis prediction.