This article documents and interrogates top-down district-wide implementation of a two-way dual language bilingual education (DLBE) program in a large urban district. We carried out a language policy ethnography to explore the way two schools' teams of third grade teachers worked together to negotiate the intersection of DLBE implementation and high stakes accountability pressures. There appears to be inherent tension between preparing children for monolingual standardized tests, and meeting DLBE program goals of bilingualism, biliteracy, high academic achievement and cross-cultural competence. The pressure to prepare children for high-stakes testing ultimately led to the dismantling of the DLBE program in both schools. Lack of training, insufficient materials and conflicting curricular mandates were further obstacles at both schools. There were important contextual differences, however; in one school, teacher agency and collaboration led to far more enriched educational experiences for children, while in the other rote test preparation took over. We recommend investment in teacher professional development and teacher agency. Ultimately, for DLBE programs to succeed, high stakes testing must give way to multiple-measure accountability that matches language program goals.