Two empirical studies conducted in a university setting evaluate Siette, a set of tools for constructing and administering adaptive tests via the Internet. Such tests guarantee reliable and accurate student knowledge diagnosis. Adaptation depends on the question-selection process (performed dynamically during the test) and the test-finalization decision (which relies on the required accuracy of the diagnoses). The studies explore how self-assessment tests influence students' final grades. This article is part of a special issue on intelligent educational systems.