A cross-continent collaboration demonstrates that neuron types in the cerebellum can be identified from extracellular recordings. Optogenetically labeled datasets, collected in Court Hull’s laboratory at Duke and Michael Häusser’s laboratory at University College London, were used to train a deep-learning classifier. This classifier utilized readily available extracellular signals, including the neuron’s action potential waveform and firing regularity statistics, as input. The classifier not only predicts the labels of withheld neurons from the ground-truth optogenetically labeled mouse database with approximately 90% accuracy but also largely agrees (>80%) with expert labels of floccular units recorded in rhesus macaques. Furthermore, we show that populations of neurons identified by the deep-learning classifier exhibit differing dynamics during common cerebellar-dependent tasks. Thus, our deep-learning classifier represents an important first step toward uncovering biological insights into the computations performed by the cerebellar circuit.