The study of social interactions in mice is used as a model for normal and pathological cognitive and emotional processes. But extracting comprehensive behavioral information from videos of interacting mice is still a challenge. We describe a computerized method and software, MiceProfiler, that uses geometrical primitives to model and track two mice without requiring any specific tagging. The program monitors a comprehensive repertoire of behavioral states and their temporal evolution, allowing the identification of key elements that trigger social contact. Using MiceProfiler we studied the role of neuronal nicotinic receptors in the establishment of social interactions and risk-prone postures. We found that the duration and type of social interactions with a conspecific evolves differently over time in mice lacking neuronal nicotinic receptors (Chrnb2-/-, here called β2(-/-)), compared to C57BL/6J mice, and identified a new type of coordinated posture, called back-to-back posture, that we rarely observed in β2(-/-) mice.