Near-field optical data are non-stationary, which means that their spectral content varies with the position of the tip, due to both the scanning-probe recording process and the variations of the optical signal. Therefore time-frequency representations are potentially powerful tools for local characterization as they distribute the energy of the analysed signal over the time and frequency variables, and faithfully depict the signal local behaviour. In this paper, the time-frequency distributions are shown to be appropriate tools to analyse near-field optical data by using it first on simulated data, and second on experimental near-field optical images. Within this context, we observe that time-frequency analysis allows a possible separation of relevant optical signals from artefacts, especially in the usual case where the near-field optical signal is lower band than the feedback data.