Motivational interviewing (MI) is a goal oriented psychotherapy that facilitates intrinsic motivation within a client in order to change behavior in a dialog setting. The Motivational Interviewing Skills Code (MISC) is a manual observational coding method used to quantify and evaluate the quality of MI sessions using their audio-visual recordings. However, this coding method is both labor intensive and expensive. We present an approach towards automating MISC assignments in MI involving addiction cure. Specifically, we focus on predicting valence for “Client Change Talk” ( ChangeTalk) utterances, which indicate a client's attitude towards a “Target Behavior Change” ( Target). We further study the effect of incorporating counselor behavior in the model. We observe that our best model achieves an unweighted accuracy of 50.8% in a 3-way classification of positive vs negative valence ChangeTalk vs no ChangeTalk. Furthermore, we study the effect of including non-verbal behavior, specifically laughters, in our model. Information regarding location of laughters improves the unweighted accuracy of our model to 51.4% and our experimental results suggest prosodic differences in laughters belonging to ChangeTalk utterances with different valences.