Collaborative learning is recognised as an essential task for assisting in the acquistion of knowledge and the development of skills. At the same time, research in learning skills has demonstrated that different learners often benefit form different approaches to the learning task, something to which an arbitrarily created collaboration can often become an obstacle, making the learning process even more difficult. faced with the task of splitting large classes into smaller groups for peer-support, educators can neither afford the necessary analysis of each student's profile nor possess the experience to use their profiles to improve the selection of groups. At the same time, using ICT to automate the task of splitting learners that will enable an algorithmic approach to modelling learners and selecting groups. In this paper we present our experiments to linking learning styles to student performance, discuss issues arising from the different methods of selecting groups for collaborative learning, contrast the and present a formal model for profiling learners showing how this model can be used to automate the selection of groups.