Emotions play a significant role in healthy cognitive functioning; they impact memory, attention, decision-making and attitude; and are therefore influential in learning and achievement. Consequently, affective diagnoses constitute an important aspect of human teacher-learner interactions motivating efforts to incorporate skills of affect perception within computer-based learning. This paper provides a discussion of the motivational and methodological issues involved in automatic affect inference in learning technologies. It draws on the recent surge of interest in studying emotions in learning, highlights available techniques for measuring emotions, and surveys recent efforts to automatically measure emotional experience in learning environments. Based on previous studies, six categories of pertinent affect states are identified; the visual modality for affect modelling is selected given the requirements of a viable measurement technique; and a bottom-up analysis approach based on context-relevant data is adopted. Finally, a dynamic emotion inference system that uses state of the art facial feature point tracking technology to encode the spatial and temporal signature of these affect states is described. (Contains 4 tables and 1 figure.)