We introduce Domain-specific Masks for Generalization, a model for improving\nboth in-domain and out-of-domain generalization performance. For domain\ngeneralization, the goal is to learn from a set of source domains to produce a\nsingle model that will best generalize to an unseen target domain. As such,\nmany prior approaches focus on learning representations which persist across\nall source domains with the assumption that these domain agnostic\nrepresentations will generalize well. However, often individual domains contain\ncharacteristics which are unique and when leveraged can significantly aid\nin-domain recognition performance. To produce a model which best generalizes to\nboth seen and unseen domains, we propose learning domain specific masks. The\nmasks are encouraged to learn a balance of domain-invariant and domain-specific\nfeatures, thus enabling a model which can benefit from the predictive power of\nspecialized features while retaining the universal applicability of\ndomain-invariant features. We demonstrate competitive performance compared to\nnaive baselines and state-of-the-art methods on both PACS and DomainNet.\n