Molecular “fingerprints” encoding structural information are the workhorse of cheminfor- matics and machine learning in drug discovery applications. However, fingerprint representa- tions necessarily emphasize particular aspects of the molecular structure while ignoring others, rather than allowing the model to make data- driven decisions. We describe molecular graph convolutions, a fully integrated machine learn- ing architecture for learning from undirected graphs, such as small molecules. Graph convo- lutions use a simple encoding of the molecular graph (atoms, bonds, distances, etc.), allowing the model to take full advantage of information in the graph structure.