Evaluating learning outcomes depends upon objective and actionable measures of what students know – that is, what can they do with what they have learned. In the context of a developmental biology course, a capstone of many molecular biology degree programs, I asked students to predict the behaviors of temporal and spatial signaling gradients. Their responses led me to consider an alternative to conventional assessments, namely a process in which students are asked to build and apply plausible explanatory mechanistic models (“PEMMs”). A salient point is not whether students' models are correct, but whether they “work” in a manner consistent with underlying scientific principles. Analyzing such models can reveal the extent to which students recognize and accurately apply relevant ideas. An emphasis on model building, analysis and revision, an authentic scientific practice, can be expected to have transformative effects on course and curricular design as well as on student engagement and learning outcomes.