Recommender Systems have found their way into many domains such as the recommendation of movies and music, or the recommendation of products in e-commerce websites. Over the last decade they have made their way into the world of Technology Enhanced Learning to recommend interesting learning materials such as novel books or informative articles. Little research has been done to investigate the collaborative classi cation power of Recommender Systems to assist students in their online learning experience. In this dissertation we use a Neighbourhood-based Collaborative Filtering algorithm to predict the diculty of online language exercises. A data-set of language exercises has been made available by Televic Education from their educational platform Edumatic, providing us with real-life data. Intuitively this means we predict the diculty of a particular exercise for a particular student based on how well similar students have performed on that exercise. Similar users are determined by a similarity measure and in this thesis we have evaluated multiple measures such as the Pearson Correlation, the Tanimoto Coecient and an extended form of the Tanimoto Coecient similarity measure. Experimental analysis shows that, when using the right combination of algorithms and similarity measures, the collaborative approach can provide an accurate classi er. The optimal combination proved to be an item-based approach with the Extended Jaccard as similarity measure. The work in thesis can be considered as part of a solution to further improve the online learning experience. By being able to predict the diculty for users on items, we can in future steps guide students towards new exercises tailored to their own needs creating an adaptive learning environment.