The following paper presents possibilities for the application of selected time-frequency analysis methods in the fault detection of cage induction machines in transient states. The starting phase current of the machine was chosen as a diagnostic signal. Selected faults were eccentricities – static and dynamic. In order to increase the selectivity of the obtained signal transformations, a notch filter was used to remove the base harmonic of the phase current. Two approaches of fault detection were compared. In the first approach, the characteristic feature of fault was extracted using DWT analysis. Next, TMCSA methodology was applied in which characteristic harmonics related to faults were shown on a time-frequency plane. In this case, applied methods were a Gabor transformation, STFT, CWT and Wigner–Ville’s transformation. In the analysis, a phase current signal approximated by DWT was used. DWT approximation was applied to filter higher harmonics which improves the resolution of the obtained transformations.