W the analog representation for jet noise studies allows the prediction of the dimensional laws for acoustic emission, a better understanding of the'acoustic far field requires a more physical picture of the acousticgenerating mechanisms. In particular, in a previous work a direct experimental estimation of the acoustic intensity associated with the large-scale structure of turbulence was obtained, on the basis of the same assumptions allowing the full use of the second-order analysis on signals recorded inside and outside the jet. Here, a more general method, based on conditional sampling, is proposed for jet noise source localization. It consists of selecting, through a detection criterion > imposed on a selected signal, a set of samples representative of the passage of wave packets, and working on an ensemble of these sets rather than on time averages. In this manner, each generating mechanism can be selected by a proper choice of the detection criterion, and its acoustic efficiency computed. In addition, ensemble averages avoid the limitations due to time averages, and lead to a better signal/noise ratio when the classical correlation function is very weak.