Shape memory polyurethanes are materials that can recover a finite pre-deformed shape in response to thermal stimuli due to the combined action of hard and soft segments in the molecular configuration. In this study, affine network models are used to describe the mechanical behaviors of the soft segment phases. Furthermore, a new four-element recovery model is constructed for the hard segment phase in consideration of the structure change during the recovery process. In addition, as shape memory polyurethanes are temperature-sensitive materials, the influence of temperature on the parameter viscosity is investigated. A phenomenological constitutive model is proposed, with which the thermomechanical behavior of shape memory polyurethanes is predicted and the simulation results agree well with the experimental results.