We study price differences based on skin tone in auctions for digital art using the complete history of online transactions for CryptoPunks, one of the earliest and most influential non-fungible tokens (NFTs). CryptoPunks are 10,000 unique digital facial images based on algorithmically generated combinations of hair, skin, eye, nose, and mouth attributes. The tokens are highly valued, with average auction sale prices of $100,000 and those composed of rare characteristics selling for several million dollars online and at traditional auction houses. Using data on the universe of CryptoPunk transactions since their launch in 2017, we illustrate significant price gaps of otherwise similar tokens based on their skin tone; tokens with black and brown skin-tone sell for approximately 10 percent less on average than light-skinned images. These price gaps are not explained by flexibly controlling for the complete set of facial features that define each image. We further explore how skin-tone price differentials change over time and vary with a token's sale frequency. The results offer an important new illustration of racial-based price gaps in a digital marketplace.