While markerless motion capture system has recently gained interest, little is known on the accuracy of Theia 3D for measuring shoulder kinematics. This study aimed to compare shoulder kinematics across the full range of arm motion between Theia 3D and a gold standard marker-based motion capture system. Shoulder kinematics of 20 healthy subjects were measured during three repetitions of bilateral full arm scaption, abduction, flexion, internal rotation, external rotation, and internal and external rotation at 90° abduction. Data were simultaneously collected with Theia 3D (v2024.1.24) markerless and marker-based motion capture. Markerless and marker-based angular trajectories had similar patterns, with larger differences towards the inflection point of the movements and with the markerless trajectories having mostly greater values than marker-based trajectories. Root mean square differences were smallest for abduction angles (<6°) and largest for the external rotation angles (15.3°–22.3°). Coefficients of multiple correlations (CMCs) were mostly good to excellent (>0.75), but CMC was moderate (0.65–0.75) for flexion angles during rotation at 90° and weak (<0.65) for flexion angles during external rotation. For almost all tasks, the ranges of motion differed significantly between the two measurement systems in all three planes. However, mean differences in the coronal plane and in the sagittal plane were within the minimal clinically important differences. We found moderate to very strong correlations between the range of motion parameters of the two measurement methods. The assessment of shoulder kinematics with Theia 3D is promising but further improvements are needed for clinical routine application.