Falls are among the major causes of injury in older adults. Although several fall detection systems exist, most are not focused on assessing the severity of a fall. OpenCap can be an important tool for this purpose, enabling ease and markerless capture of the 3D kinematics of falls. However, OpenCap has only been validated on a limited range of exercises and movements, and its applicability to fall detection or analysis remains uncertain. This study aims to evaluate the performance of OpenCap in capturing falls, using the XSens inertial motion capture system as a comparison. Five healthy volunteers were recorded with the OpenCap and XSens systems while performing two types of falls (i.e., forward and backward) onto mattresses. Each fall type was repeated five times per participant. Errors in 3D joint kinematics were quantified using the Mean Absolute Error across 24 rotational degrees of freedom and three translational degrees of freedom. The results indicate that OpenCap can capture most of the fall kinematics patterns with a global rotational error of \(10.70^\circ \) for frontal falls and \(10.85^\circ \) for backward falls, and a translational error of 8.0 cm and 8.4 cm, respectively. A marked difference between upper and lower body rotational MAEs was observed: for frontal falls, the MAE was \(13.90^\circ \) for the upper body and \(7.88^\circ \) for the lower body; for backward falls, the corresponding values were \(12.77^\circ \) and \(8.86^\circ \) .