Affective components are frequently overlooked in computational modelling, despite the notable role of emotions in learn-ing. Towards the goal of measuring affect in learning, we developed a theory-based Bayesian model that predicts surprisebased on a learners prior beliefs and the evidence observed, and then compared the model to a physiological measure com-monly suggested to capture surprise: pupil dilation. Critically, we also investigate whether this correlation is strong whenparticipants predict the events. Comparing our model predictions to the first four test trial responses from 93 participants(mean age: 8.00 years) revealed a significant, positive correlation when making predictions (r(9)=.55, p=0.04), a negativecorrelation when only evaluating (r(9)=-.50, p=0.07), and significant difference between groups (z=2.34, p¡0.01). Nextsteps will allow us to build on this result by developing a modified Bayesian model, that takes physiological surprise as acomponent in predicting the participants learning.