This article examines the conditions under which social learning occurs and leads to adaptive measures through two empirical examples of Peruvian cities that invested in watershed protection for their urban water supplies. Social learning is an increasingly popular approach aimed at achieving socio-ecological resiliency through multi-stakeholder collaborative governance processes. Social learning is a convergence in knowledge that occurs through dialog and deliberation. Yet, assumptions that social learning will necessarily lead to more environmentally sustainable and resilient practices may be overly optimistic, especially as they rarely consider the political and organizational dimensions of decision making. This study analyzes two seemingly similar case studies of multi-stakeholder water management in Peru that resulted in watershed protection programs—a novelty in Peru that will help ensure future water supplies. Despite similar programs adopted, though, the social interactions were markedly different. Social learning occurred in Moyobamba, where the multi-stakeholder platform was characterized by trust, flexibility, and sustainability. In Cusco, however, stakeholders reached an agreement on projects for watershed protection, but the process exhibited little evidence of social learning, trust, or flexibility. In this article, I use process tracing to analyze if and how social learning occurred in each case. Then, I identify factors that contributed to social learning, including diverse participation, open communication, multiple sources of knowledge, extended engagement, unbiased facilitation, and an opportunity to influence outcomes.