To avoid inconvenience caused by shaft rotation and improve immunity to measurement noises, an elaborately designed scheme is proposed for parameter identification of induction motors. After analyzing the responses of a simple step-voltage test, a sequence of pseudorandom signals, customized to excite abundant dynamics in the featured frequency band of the motor, are injected into the stator in a single-phase mode at standstill. The crucial feature of the proposed scheme is that a nonlinear procedure is introduced to minimize "predicted errors" of the estimation model, which lowers influences of measurement noises notably, and thus the design of low-pass filters is simplified greatly. Experimental comparisons are carried out, including not only tests on a squirrel-cage motor, but also extended tests on a wound-rotor motor to testify accuracy of rotor-side parameters, both using a real inverter. The results indicate that the proposed scheme is able to estimate parameters required by controllers accurately in the noisy environment, and improve performances of sensorless motor drivers.