A novel measure for automatically quantifying the amount of interpersonal influence present in face-toface conversations is proposed based on the visualattention patterns of the participants as inferred from video sequences. First, we focus on the gaze of the participants as an indicator of addressing / listening behavior and build a probabilistic conversation model for inferring the gaze directions and conversation structures like monologue and dialogue, from observed utterances and head directions measured with imagebased head trackers. Next, based on the estimates, the amount of influence is defined based on the amount of attention paid to speakers in monologues and to persons with whom the participants interact with during the dialogues. Experiments confirm that the proposed measures reveal some aspects of interpersonal influence in conversations. Keywords Visual attention, eye gaze, conversation, influence, dynamic Bayesian network, Markov chain Monte Carlo ACM Classification Key...