Rehabilitation robots have significant potential to facilitate the recovery of lost upper extremity function following stroke. However, they have not produced better functional outcomes than those achieved through conventional therapy, in part because generic robot control algorithms (e.g., assist as needed, error augmentation) do not take into account patient-specific neural control deficiencies. One way to address this limitation is by combining upper extremity neuromusculoskeletal models with rehabilitation robot models, thereby permitting the design of patient-specific robot control algorithms. As a first step toward this goal, this thesis addresses the challenges involved in combining an upper extremity musculoskeletal model with a rehabilitation robot model. The development of the combined arm-robot model consisted of building a model for the two DOF Kinarm rehabilitation robot and coupling that model with a published upper extremity neuromusculoskeletal model. This process was complex due to the many instances of closed kinematic chains, therefore simulations for verification of both developed models and validation against experimental data were a pivotal focus of this study. Experimental data were collected from the Kinarm robot with and without a subject for 12 different planar motions of the arm. Verification was performed on four different configurations of models and controllers: robot model, robot controlled arm-robot model, arm controlled arm-robot model, and cooperative controlled arm-robot model. In addition, both the robot model and combined model were validated against experimental data. All four configurations were verified to reproduce experimental motion with high levels of accuracy and both models were validated to accurately recreate experimental torques with improvement possible the in arm-robot model. The coupled arm-robot model presented in this thesis can serve as the foundation for development of cooperative arm-robot control algorithms.