The academic e-learning practice has to deal with various participation patterns and types of online learners with different support needs. The online instructors are challenged to recognize these and react accordingly. Among the participation patterns, special attention is requested by dropouts, which can perturbate online collaboration. Therefore we are in search of a method of early identification of participation patterns and prediction of dropouts. To do this, we use a quantitative view of participation that takes into account only observable variables. On this background we identify in a field study the participation indicators that are relevant for the course completion, i.e. produce significant differences between the completion and dropout sub-groups. Further we identify through cluster analysis four participation patterns with different support needs. One of them is the dropout cluster that could be predicted with an accuracy of nearly 80%. As a practical consequence, this study recommends a simple, easy-to-implement prediction method for dropouts, which can improve online teaching. As a theoretical consequence, we underline the role of the course didactics for the definition of participation, and call for refining previous attrition models.