A pattern recognition technique based on Bayes minimum error classifier is developed to detect broken rotor bar faults in induction motors at the steady state. The proposed algorithm uses only stator currents as input without the need for any other variables. First rotor speed is estimated from the stator currents, then appropriate features are extracted. The produced feature vector is normalized and fed to the trained classifier to see if motor is healthy or has broken bar faults. Only number of poles and rotor slots are needed as pre-knowledge information. Theoretical approach together with experimental results derived from a 3 hp ac induction motor show the strength of the proposed method. In order to cover many different motor load conditions data are obtained from 10% to 130% of the rated load for both a healthy induction motor and an induction motor with a rotor having 4 broken bars.