This paper investigates the application of Unscented Kalman Filter (UKF) for Induction Motor (IM) sensorless drives. UKF's use nonlinear Unscented Transforms (UT) in the prediction step in order to preserve the stochastic characteristics of a nonlinear system. The advantage of using Uts is their ability to capture the nonlinear behavior of the system, unlike extended Kalman filters that use linearized models. Four original variants of the UKF for IM state estimation, based on different UTs are described, analyzed, and compared. The four transforms are basic, general, simplex, and spherical UTs. This paper discusses the theoretical aspects and implementation details of the four UKFs. It is concluded that the UKF is a viable and powerful tool for IM state estimation and that basic and general UTs give more accurate results than simplex and spherical UTs.