The work carried out by the authors consists of applying a modern time-frequency decomposition (TFD) tool, the Hilbert-Huang Transform (HHT), to the diagnosis and the evaluation of electromechanical faults in induction machines. These machines are widely spread nowadays, being involved in many industrial processes as well as in power generation installations such as nuclear plants. The core of the proposed methodology is the analysis of the current demanded by the stator winding of the machine during its connection process known as startup transient. Once the current is analyzed, characteristic patterns caused by the evolution of certain components created by the corresponding faults are identified; this evolution is due to the dependence of these fault-related components on the slip s, a quantity varying during a direct startup transient from 1 to near 0. In the present paper, the HHT is applied to the diagnosis of two different faults: rotor bar breakages and mixed eccentricities. In comparison with other TFD tools, the HHT provides certain advantages that are discussed in the work. The validity of the approach is proven through several experimental tests on real machines with different sizes and characteristics. The results show the potential of the methodology for reliable fault diagnosis and for correct discrimination between the different electromechanical failures.