Stator winding faults may cause severe damages in Permanent Magnet Synchronous Motors (PMSM) if not detected early on. The earliest fault detection in motors should be made during transient states throughout the initial starting period. A new approach based on Empirical Mode Decomposition (EMD) and statistical analysis was presented for detecting stator winding fault by way of transient state phase current of PMSM in this study. Models based on finite elements method were developed for the PMSM representing the healthy and faulty states in order to implement the suggested fault detection method. Afterwards, transient state stator phase winding currents were measured for healthy and faulty states under nominal load in accordance with motor models. These non-linear current signals monitored were separated into its Intrinsic Mode Functions (IMF) via the EMD method. Pearson Correlation Coefficient was used for determining the IMF that most resembles the characteristics of the main signal. Statistical parameter-based feature extractions were carried out for the IMF signals determined for the healthy and faulty states. Fault and fault level detection were carried out successfully by comparing the obtained feature vectors. The acquired results have put forth that the suggested method can be used securely for fault detection in electrical machines especially for early fault detection.