Online quizzes constitute a classical building block of digital learning environments. Their answers can be aggregated using Bayesian Knowledge Tracing (BKT), a widely used model for student knowledge evaluation. But traditional quiz systems force the user to provide precise answers, limiting the expression of her doubt, uncertainty or imprecision. To address these limitations, we introduce a novel approach called Credal Knowledge Tracing (CrKT), based on the Dempster-Shafer Theory. In this setting, a doubting learner is not constrained to choose randomly an answer but can choose different ones simultaneously, and can weight each of them with a level of certainty. In this work, we provide a definition of CrKT and compare it with alternative approaches. Using a multi-attempt simulation framework, our experiments demonstrate that CrKT converges more rapidly to knowledge mastery than BKT, highlighting CrKT’s potential to enhance intelligent tutoring systems.