Quasi-periodic oscillations and dwarf nova oscillations occur in dwarf novae and nova-like variables during outburst and occasionally during quiescence, and have analogues in high-mass X-ray binaries and black-hole candidates. The frequent low coherence of quasi-period oscillations and dwarf nova oscillations can make detection with standard time-series tools such as periodograms problematic. This paper develops tools to analyse quasi-periodic brightness oscillations. We review the use of time-frequency representations in the astronomical literature, and show that representations such as the Choi-Williams Distribution and Zhao-Atlas-Marks Representation, which are best suited to high signal-to-noise data, cannot be assumed a priori to be the best techniques for our data, which have a much higher noise level and lower coherence. This leads us to a detailed analysis of the time-frequency resolution and statistical properties of six time-frequency representations. We conclude that the wavelet scalogram, with the addition of wavelet ridges and maxima points, is the most effective time-frequency representation for analysing quasi-periodicities in low signal-to-noise data, as it has high time-frequency resolution, and is a minimum variance estimator. We use the wavelet ridges method to re-analyse archival data from VW Hyi, and find 62 new QPOs and 7 new long-period DNOs. Relative to previous analyses, our method substantially improves the detection rate for QPOs.