This paper presents an algorithm for three-phase induction motors rotor flux observation based on a neofuzzy neuron (NFN) network with real-time training featuring quick accurate convergence, great adaptability to system dynamics, good performance over a wide speed range, and low sensitivity to parameter variations (rotor and mutual inductances and rotor resistance) and disturbances. The observer requires only rotor speed and stator current measurements as input signals and. its fuzzy-neural network does not need previous training. Convergence in one step is demonstrated and the influence of the learning rate on performance is analyzed. Both digital simulation and experimental results are presented and show the observer good overall performance.