The intricate motion of the small bones of the feet are critical for its diverse function. Accurately measuring the 3-dimensional (3D) motion of these bones has attracted much attention over the years and until recently, was limited to invasive techniques or quantification of functional segments using multi-segment foot models. Biplanar videoradiography and model-based scientific rotoscoping offers an exciting alternative that allows us to focus on the intricate motion of individual bones in the foot. However, scientific rotoscoping, the process of rotating and translating a 3D bone model so that it aligns with the captured x-ray images, is either semi- or completely manual and it is unknown how much human error affects tracking results. Thus, the aim of this study was to quantify the inter- and intra-operator reliability of manually rotoscoping in vivo bone motion of the tibia, talus, and calcaneus during running. Three-dimensional CT bone volumes and high-speed biplanar videoradiography images of the foot were acquired on six participants. The six-degree-of-freedom motions of the tibia, talus, and calcaneus were determined using a manual markerless registration algorithm. Two operators performed the tracking, and additionally, the first operator re-tracked all bones, to test for intra-operator effects. Mean RMS errors were 1.86 mm and 1.90° for intra-operator comparisons and 2.30 mm and 2.60° for inter-operator comparisons across all bones and planes. The moderate to strong similarity values indicate that tracking bones and joint kinematics between sessions and operators is reliable for running. These errors are likely acceptable for defining gross joint angles. However, this magnitude of error may limit the capacity to perform advanced analyses of joint interactions, particularly those that require precise (sub-millimeter) estimates of bone position and orientation. Optimizing the view and image quality of the biplanar videoradiography system as well as the automated tracking algorithms for rotoscoping bones in the foot are required to reduce these errors and the time burden associated with the manual processing.