Recent developments in cognitive and educational psychology, such as increased appreciation of the situated nature of learning and understanding, call for broader ranges of student models and types of data than those standard in testing today. We must specify how what we observe on the test is related to competence as we conceptualize it, and construct a framework for carrying out inferences. There is a growing need for testing that verifies performance within a given situation or context, or in the completion of a complex task requiring the language learner to integrate several aspects of learning. A broader view of testing to include these issues requires clear definition of what is to be measured, new methods of eliciting evidence, and methods of weighting various sources of evidence for decision-making about test results. For the latter, it is possible to analyze evidence statistically so the mathematical models mirror the inference model based on that evidence. A conceptualization of test theory designed to address issues of weight and coverage of evidence for statements framed in more recent educational/psychological paradigms is outlined, with examples. (Contains 26 references.) (Author/MSE)