This paper aims to describe a speed sensorless control of induction motor (IM) based on a new neural network estimator of the rotor speed. This method is based on the extraction of the rotor speed information from the command frequency and the phase difference between stator current and voltage. The data used to train the artificial neural network (ANN) are obtained from steady state equivalent circuit of the induction motor simulated for various frequency. The experimental results show the efficiency of the proposed speed estimation technique at high speed and at very low speed.