There has been an increasing number of quantitative methods to measure and evaluate player experiences. However, current methods either focus on telemetry approaches, which are insufficient to capture real life responses, or psychophysiological methods, which are intrusive and more suited to controlled laboratory environments. This paper presents the position that computer vision techniques can provide a less intrusive and more versatile solution for automatic evaluation of game user experiences. A conceptual framework to automatically infer flow intensity is presented and a work-in-progress study is included to demonstrate the feasibility of this research direction.