When we rely upon gains on some measure to support statements of prescription, we have the obligation to ensure that those measures are valid. Nearly 10 years after an influential National Research Council ("2001") report on educational assessment identified an explicit model of cognition as one of three necessary components of a valid assessment system, we note that most measures still lack this fundamental cornerstone. In this paper, we draw attention to the "construct modeling" approach to assessment that strives for coherence and consistency with a model of cognition in which student proficiency varies along a continuum of competence. This approach is illustrated in the context of an assessment of conceptual understanding of certain scientific phenomena given to undergraduates at a large public university (National Research Council "2001").