Fish size distributions are informative indicators of ecological and fisheries status, yet obtaining accurate measurements underwater poses challenges. Traditional capture methods are laborious and prone to bias, while stereoscopic imaging struggles with automation due to underwater background uniformity. To address these issues, we first engineered a stereo correspondence matching algorithm and developed software to measure fish lengths with custom stereoscopic camera rigs. Second, we adapted a deep learning model to estimate distance from camera and size using monocular imagery. We measured performance against the range and size ground truths for thousands of fish across diverse genera and marine habitats. The more traditional stereoscopic and the novel monocular methods both achieved very good accuracy and precision. Fish sizes were estimated by the stereoscopic method within 2-9% of ground truth sizes, and the monoscopic method within 4-12%. The approach with monocular cameras and automated sizing streamlines and simplifies methods for size estimation, and given the precision achieved, offers a practical solution for enhancing underwater fish sizing, and unleashing the potential for other fish biometrics.