In this paper, the exponentially tracking control problem for a magnetic levitation system (MLS) in the presence of parameter uncertainties and external disturbances is investigated. The disturbance/uncertainties-rejecting problem for the MLS is addressed from the view of a continuous nonlinear robust control development. Another problem of the concern is the common unidirectional input constraint in the MLS (i.e., the control input, often referring to the square of the electric current, should be nonnegative during the operation of the system). By utilizing an input transformation and augmented dynamics, a virtual control input, which removes the unidirectional constraint, is affinely emerged in the dynamics. A second-order filter is introduced to provide feedback signals for the control development. A novel Lyapunov function guarantees the exponentially tracking stability of the close-loop dynamics. Finally, some numerical simulation and real-time experimental results for tracking of time-varying trajectories are presented to validate the performance of the proposed control design.