In this paper we compare a variety of non-parametric time-frequency methods to determine the best time-frequency representation (TFR) for a collection of signals. These methods include quadratic time-frequency transforms, atomic decomposition and adaptive quadratic time-frequency transforms. The performance measures used to assess the TFRs include; two-dimensional correlation, IF correlation and time-frequency resolution. Synthetic signals with different time-frequency characteristics were used in the comparison to show the strengths and weaknesses of the different time-frequency methods. It was determined that adaptive quadratic time-frequency representations provide the best overall performance and should be used if no a priori information about the time-frequency characteristics of a signal is known.