How do we measure knowledge in use? In this paper we describe how we use principles of evidence-centered design to develop classroom-based science assessments that integrate three dimensions of science proficiency—disciplinary core ideas, science practices, and crosscutting concepts. In our design process, we first elaborate on, or “unpack”, the assessable components of the three dimensions. We then use these elaborations to specify a set of claims called learning performances that describe what students need to be able to know and do in order to meet knowledge-in-use learning goals, such as the performance expectations articulated in the U.S. Next Generation Science Standards. Learning performances are crafted as knowledge-in-use statements that integrate aspects of the three dimensions, but are smaller in scope than endof-grade-band performance expectations. Next, we define task features to elicit from students the desired evidence of proficiency. Our final step entails using design patterns derived from specifying learning performances, specifying evidence, and defining task features to construct tasks that measure science proficiency. We present our design approach, provide examples of tasks, and consider implications of this work for next generation science assessment.