Studies on the psychosensory pupil response often carefully control the lighting conditions in the experiment or require a calibration procedure for each subject under different light conditions for a baseline which is later used to attenuate the pupil light response (PLR) effects from the pupil using steady state models, disregarding the dynamic nature of the pupil. Such approaches are not feasible “in the wild” since they require carefully controlled experimental conditions. We address these shortcomings in the context of screen viewing in a dataset containing 140 subjects playing a first person shooter video game and use an existing dynamic PLR model to attenuate the effects of luminance. We compute the perceived luminance using the pixel values of the screen and show that using the dynamic PLR model is more effective in attenuating the effects of luminance compared to steady state models. Subsequently, we show that attenuating the PLR from the pupil size data improves the performance of machine learning models trained to predict arousing game events compared to using the pupil size without attenuating the PLR. The implications are that our approach for estimating the perceived luminance and attenuating its effects from the pupil data can be applied to screen viewing (including VR) to unobtrusively and continuously monitor users’ emotional arousal via the pupil size.