1) The document presents a method for recognizing American Sign Language (ASL) fingerspelling using convolutional neural networks (CNNs) applied to depth maps of the hand. 2) CNNs are trained on a dataset of 1000 images of 29 different ASL alphabet signs collected from multiple subjects. The CNNs achieve 99.9% accuracy on recognized signers and 83.58-85.49% accuracy on new signers. 3) The method segments the hand from the depth image and extracts features using CNNs. It is tested in different training modes and achieves real-time recognition of 3ms per image while maintaining high accuracy.