This study aimed to investigate the reproducibility of pupil size measurements over time and between reading methods when comparing human-assisted reading to automated reading. Pupillary data were analyzed on a subset of myopic children enrolled in a multicenter randomized clinical trial on myopia control with low-dose atropine. Pupil size measurements were obtained prior to randomization at two time points (screening and baseline visits) using a dedicated pupillometer under mesopic and photopic conditions. A customized algorithm was built to perform automated readings, allowing comparisons between human-assisted and automated readings. Reproducibility analyses followed the principles of Bland and Altman and included the calculation of the mean difference between measurements and limits of agreement (LOA). We included 43 children. Mean (standard deviation) age was 9.8 (1.7) years and 25 (58%) children were girls. Using human-assisted readings, reproducibility over time showed mesopic mean difference of 0.02 mm with LOA from −0.87 mm to 0.91 mm, whereas photopic mean difference was −0.01 mm with LOA from −0.25 mm to 0.23 mm. Reproducibility between human-assisted and automated readings was also higher under photopic conditions, with mean difference of 0.03 mm and LOA from −0.03 mm to 0.10 mm at screening and mean difference of 0.03 mm and LOA from −0.06 mm to 0.12 mm at baseline. Using a dedicated pupillometer, we found that examinations performed under photopic conditions demonstrated higher reproducibility over time and between reading methods. We speculate whether mesopic measurements are sufficiently reproducible to be monitored over time. Furthermore, photopic measurements may be of greater relevance when evaluating the side effects of atropine treatment, such as photophobia.