Circular orientation cages have been used for several decades to record the migratory orientation of passerine migrants, and have been central to the investigation of the functional characteristics of the biological compasses used for orientation. The use of these cages offers unique possibilities to study the migratory behaviour of songbirds, but suffers from statistical limitations in evaluating the directions of the activity recorded in the cages. The migratory activity has been reported to vary, including complex multimodal orientation of migratory passerines tested in orientation cages irrespective of species studied. The currently applied circular statistical methods fail to describe orientation responses differing from unimodal and axial distributions. We propose for the first time a modelling procedure enabling the analysis of multimodal distributions at either an individual or a group level. In this paper we compare the results of conventional methods and the recommended modelling approach. Migratory routes may be more complex than a simple migratory direction, and multimodal behaviour in migratory species at the individual and population levels can be advantageous. Individuals may select the expected migratory direction, but may also return to safer sites en route, i.e. sites already known, which provide food and/or shelter in reverse directions. In individual birds, several directions may be expressed in the same test hour. At the species level, multimodal orientation may give an opportunity to expand the range or may refer to differential migration route preferences in different populations of birds. A conflicting experimental situation may also result in a different preferential orientation. In this paper we suggest a statistical solution to deal with these types of variations in orientation preference.