Purpose: 4D CT allows for the visualisation of joint motion in musculoskeletal applications. Quantitative analysis of joint kinematics requires postprocessing of data including registration. Introducing a bone mask (segmentation) improves registration results and helps to obtain individual bone deformations which facilitates estimation of kinematics. However, manual bone segmentation of CT time series is labour intensive and hinders application in clinical routine. We propose an automated workflow for the piece-wise rigid registration of bony structures by introducing a multi-atlas segmentation approach in the dynamic registration process. We evaluated our registration workflow on dynamic sequences of the thumb base and the knee joint to obtain metrics that describe joint kinematics. Materials and Method: In this IRB approved prospective study, 15 healthy volunteers received dynamic CT scans (256 slice Revolution CT, GE Healthcare) while performing instructed cyclic joint movements: opposition-reposition movement of the thumb (5 volunteers) and flexion-extension of the knee (10 volunteers). Reference images were selected from each dataset and automatically segmented using a multi-atlas segmentation (MAS) approach. A sequential pairwise intensity-based registration between the reference image and subsequent time points (moving images) was implemented... The series of rigid transformation matrices (Tbone) obtained for each bone of interest and for each time point were used to estimate cardan angles which describe the motion of each bone. Target Registration Error (TRE) was used to compare our automated registration workflow to an approach with expert manual segmentation. TRE was computed as the distance between landmarks identified on the moving images and landmarks on the reference images transformed using results of the registration. Results: For the knee data, the tibiofemoral joint reached 1.3˚ [95%CI: 2.3-0.41] of adduction and 5.9˚ [95%CI:6.9-4.8] internal rotation of the tibia relative to the femur in the first 30 degrees of knee flexion. 50.6˚ [95%CI: 57.6-43.7] of flexion, 12˚ [95%CI: 7.0-18.7] of abduction and 23˚ [95%CI: 15.5-32.2] internal rotation was calculated for the movement of first metacarpal. No significant difference in TRE was observed between our proposed approach and results obtained using expert manually segmented images (p=0.51). (Figure Presented) Fig. 1 (abstract OL29). 3D kinematics estimated using our dynamic registration workflow of the (a) knee (b) thumb. Conclusions: We setup an automated workflow for the non-invasive estimation of three-dimensional in vivo joint kinematics based on dynamic musculoskeletal CT images. Our work contributes to an automated workflow, and hence clinical feasibility, for a quantitative motion analysis of dynamic CT MSK data.