Active magnetic bearings (AMBs) offer contact-free and frictionless support of rotating machinery. However, because of their limited force capacity, they have to incorporate retainer bearings to protect the rotor and stator laminations against high-amplitude vibration levels. Efficient modeling of contact dynamics is important for the design of adaptive controllers to prevent contact. If, however, contact does occur, it is necessary to recover the rotor position with minimum damage and without shutting down the system. This paper utilizes constrained Lagrangian equations of motion to develop a computationally efficient method to model contact dynamics. The method does not require a direct physical modeling of contact forces, although the contact forces are automatically evaluated from the constraint conditions, and it can be applied to multicontact cases. Furthermore, the technique is capable of detecting and simulating the destructive backward whirl rolling motion. A model reduction technique is introduced to improve the computational efficiency. This is demonstrated by comparing numerical predictions with experimental results, obtained for a 2-m-long flexible rotor supported by two magnetic bearings.