The fact that toehold strategies are rarely used in recent acquisitions is inconsistent with their claimed advantages, and has been suggested to constitute a ‘toehold puzzle’. We present a learning explanation for the toehold puzzle showing that the more cautious usage of toeholds is improving toehold performance. First, when an IV regression controls for the more difficult deal characteristics of toehold deals, we show that toeholds do increase returns to acquirers. Second, toehold performance improves over time, which can be attributed to the more effective acquisition of the toeholds over time. Third, we find that experience of toehold bidding helps acquirers to incorporate optimal toehold threshold considerations better. (Coauthors S. Gryglewicz, H. Smit.)