Abstract. We present results from a pilot study on the utility of automatic facial expression recognition for inferring important state about the individual student. In particular, we show how facial expression can be effectively used to estimate the level of difficulty, as perceived by an individual student, of a delivered lecture. We also show how facial expression data can be used to predict the individual student's preferred rate of curriculum presentation at each moment in time. On a recorded video lecture viewing task, training on less than two minutes of recorded facial expression data (half the total lecture length), our system predicted the subjects' self-reported difficulty scores with mean accuracy of 42% (Pearson correlation) and the subjects' preferred viewing speed with mean accuracy of 29%. Our techniques are fully automatic and have potential applications both for intelligent computer teaching aids as well as for teaching by humans in standard classroom environments.