The mechanisms involved in the generation of heart sounds have always been of interest, mainly for diagnosis purposes. As a result, mathematical models have been proposed for first (S1) and second (S2) heart sounds and different efforts have been made to select the best signal processing tool to analyze them. Different frequency analysis techniques have been used to relate cardiac structure to the vibration they emit. In this work, we applied the empirical mode decomposition (EMD), a recently developed technique, for time-frequency (TF) analysis of heart sounds. EMD has shown interesting properties for biomedical signals related to nonlinear and non-stationary analysis. EMD is an adaptive decomposition since the extracted information is obtained directly from the signal without the use of kernels or mother waveforms. In this paper, EMD is first investigated in simulated scenarios through mathematical models for SI and S2 to validate its performance. Later, a real heart sound acquired over the thoracic surface of a healthy subject is analyzed. The work points out the advantage of EMD for this task.