This study investigates wound rotor induction machine bearing fault detection by stator current analysis. The research first establishes an analytic machine model that enables high fidelity simulation of a range of machine bearing defects. The timestepped simulation results are then used to examine stator current spectral signatures of typical bearing faults. The calculations indicate that a number of low magnitude fault specific frequency components appear in the current signal as a result of air-gap variations produced by an incipient bearing fault. However, the considerably low magnitude levels at which these components are exhibited would make the detection of bearing fault using conventional current signature analysis techniques challenging. An alternative technique based on spectral analysis of complex current signals is therefore proposed in order to improve fault detection. The validity of the findings of this work is confirmed by analysis of measured data obtained on a 30 kW commercial machine test rig that can be configured to introduce a range of different bearing fault severities. © 2013 The Institution of Engineering and Technology.