Context-aware ubiquitous learning environment, which embeds various ubiquitous computing technologies, allows applications to acquire diverse learning behaviors of each u-learner. Those behaviors that were collected and recorded by the u-learning system can be quite useful to enhance the analysis of learner characteristics which can be utilized to distinguish and group learners for further instruction strategy design, for instance, a ubiquitous collaborative learning strategy. This needs a systematical method to analyze learner behaviors and utilize learner characteristics for the group composition. This paper attempts to propose an effective learner grouping scheme which contains processes of the transformation from uportfolio to our proposed Portfolio Grid, the creation of learner similarity matrix, and the group composition. The aim of this paper is to propose a systematical manner of dividing learners into groups based on the analysis of previously collected u-learning portfolios. The analysis outcome, which has implied the similarity of various behaviors of learners, will facilitate obtaining an expected grouping result for the use of further learning activities such as ubiquitous collaborative learning and team working.