Roller bearings are important elements of high performance mechanical systems, as they support the rotating parts and they reduce the contact friction. Therefore, it is necessary to continuously monitor their performance using non-destructive diagnostic methods. These methods interrogate data signals acquired during the machines operation. In this paper, the method of Hilbert – Huang transform was applied on non-stationary and non-linear vibration signals obtained via accelerometers from a mechanical system which contains a shaft rotating on Roller bearings. Two series of signals were used. The first series obtained from healthy bearings. The second series, obtained from bearings with seeded defects. The experimental study and its results are presented, as well as the conclusions drawn from the results.