In this study, state estimators are designed and compared for surface-mounted permanent magnet synchronous machine (PMSM) drives. The state-space representations used for estimator design are the infinite inertia model, the full model including the equation of motion, and an augmented description extended by the permanent magnet (PM) flux linkage as an additional state variable. By using the latter model, the estimators are less sensitive to the PM flux linkage mismatches due to the online identification of this parameter. For each nonlinear models, observability analyses are presented with and without position measurement. Based on all six models, state estimators are implemented using unscented Kalman filter (UKF) algorithm, and the estimation performances are compared by simulations.