Abstract: In this work, we study transfer learning for supervised learning. In particular, we propose and analyze some probabilistic measures of transferability, which we refer to as transfer scores. We then demonstrate the relevance between the transfer scores and transfer learning performances via various numerical examples.