Short-circuit faults in windings due to the deterioration of insulation are one of the most common faults in motor drive systems. An easy and effective fault diagnostic method is urgently required to ensure the highly reliable operation. This paper proposes a novel online diagnosis method for short-circuit faults in the stator winding inside a low-voltage induction motor running under variable load conditions. In this study, first, several motor operations are carried out under various load conditions, and the features that represent the characteristics corresponding to the condition of the motor winding are extracted. From this experiment, it is observed that these features are linearly distributed in three-dimensional space. Second, some straight lines approximating this feature distribution obtained from healthy motors are derived, and the distances between the straight lines and the features obtained from a target motor are calculated. Then, we introduce a fault probability that can be used to diagnose the condition of the winding. On the basis of this probability, the quality of the stator winding is determined. The effectiveness of the proposed diagnosis method is experimentally verified.