The number of tutoring systems among e-learning educational systems in all domains has been steadily increasing. Such systems are designed with the goal of taking over the role of the real teacher who is otherwise responsible for all aspects of the teaching process, by autonomously applying the same or very similar methods to those of an expert in a particular field. In these learning contexts it is necessary to ensure a personalised approach to teaching due to unavoidable differences in learner characteristics. System users usually make a more or less heterogeneous group in which individuals differ in their features such as the level of knowledge or skill, previous experience, learning style, goals or interests, to name just a few. Improper treatment of learner differences may lead to a drop in motivation regarding system use or even a complete failure of the educational setup. That is why it is crucial to enrich the educational tutoring software with adaptive features. However, introducing adaptivity into an educational system increases demands concerning its design as it is necessary first to identify user characteristics, and then ensure continuous tracking of her work, analyse her interactions with the system, manage data about individual characteristics and the ways of updating them, enable reasoning about teaching scenarios and the delivery of subsequent activity, etc. The complexity of implementation is increased in the case of the so called ill-defined domains where it is particularly challenging to structure expert knowledge in a meaningful way. One such domain is language learning: a domain not only based on knowledge but skill as well, which contributes significantly to the difficulty of its implementation. This doctoral thesis describes an adaptive e-learning system for English language reading called LLS, which considers the level of a user's linguistic skill estimated according to the Common European framework of reference for languages (CEFR) and guides her towards higher levels of the standard. The system's primary task is to collect initial knowledge about the user by employing an adaptive test procedure, so as to be able to start with (meaningful) guidance as soon as possible. Guiding students while learning is implemented using recommendations of activity types which are, given the state and level of user's knowledge and competence, considered as optimal at that particular time, and delivery of materials of appropriate difficulty. The thesis further validates the way user data is initially obtained and discusses the effects of giudance on the overall achievement.