A systematic procedure for modeling and robust control of a multivariable magnetic levitation system is described. Our previous study revealed that an observer‐based LQ controller can stabilize the system, but generates spillovers in the presence of an impulse disturbance. To solve this problem, we apply an H∞ control to suppress the spillovers caused by unmodeled dynamics which we estimate using closed loop identification. First, an exactly linearized model is obtained to compensate for nonlinearities in the system, followed by estimation of the unmodeled dynamics using closed loop identification. Second, this information is used to design a two degree of freedom H∞ servo system for suppressing the spillover while tracking a step reference input. Finally, the desired robust performance of the resulting servo system is confirmed theoretically by γ‐analysis and also experimentally.