This article reports on the collaboration of six states to study how simulation-based science assessments can become transformative components of multi-level, balanced state science assessment systems. The project studied the psychometric quality, feasibility, and utility of simulation-based science assessments designed to serve formative purposes during a unit and to provide summative evidence of end-of-unit proficiencies. The frameworks of evidence-centered assessment design and model-based learning shaped the specifications for the assessments. The simulations provided the three most common forms of accommodations in state testing programs: audio recording of text, screen magnification, and support for extended time. The SimScientists program at WestEd developed simulation-based, curriculum-embedded, and unit benchmark assessments for two middle school topics, Ecosystems and Force & Motion. These were field-tested in three states. Data included student characteristics, responses to the assessments, cognitive labs, classroom observations, and teacher surveys and interviews. UCLA CRESST conducted an evaluation of the implementation. Feasibility and utility were examined in classroom observations, teacher surveys and interviews, and by the six-state Design Panel. Technical quality data included AAAS reviews of the items' alignment with standards and quality of the science, cognitive labs, and assessment data. Student data were analyzed using multidimensional Item Response Theory (IRT) methods. IRT analyses demonstrated the high psychometric quality (reliability and validity) of the assessments and their discrimination between content knowledge and inquiry practices. Students performed better on the interactive, simulation-based assessments than on the static, conventional items in the posttest. Importantly, gaps between performance of the general population and English language learners and students with disabilities were considerably smaller on the simulation-based assessments than on the posttests. The Design Panel participated in development of two models for integrating science simulations into a balanced state science assessment system. (Contains 5 figures and 2 tables.)