Creating accurate haptic models of physical devices is critical to improving the realism and immersiveness of virtual environments. This paper addresses the problem of modeling the feel of an automotive turnsignal switch, a prototypical device chosen for its strongly nonlinear dynamic behavior. Parameters of a position- and direction-dependent nonlinear impedance model are estimated from experimental data using a modified version of the exponentiallyweighted least-squares (EWLS) algorithm. The method is then validated through playback experiments in which subjects rate the realism of the resulting models.