This article describes challenges and benefits of modeling learner variability in Computer-Assisted Language Learning. We discuss the learner model of "E-Tutor," a learner model that addresses learner variability by focusing on certain aspects and/or features of the learner's interlanguage. Moreover, we introduce the concept of phrase descriptors, the means by which the student model of "E-Tutor" captures very detailed linguistic information on the learner's performance and progress. Finally, we provide longitudinal data that emphasize the importance of monitoring fine-grained information and underline the dynamism and non-linearity of the SLA process, as also described by Dynamic Systems Theory (DST). (Contains 7 figures, 2 tables and 12 notes.)