Volcanoes produce widely varying seismic signals due to the presence of complex and non-linear physical processes. The temporal distribution of seismicity at volcanoes ranges from individual transients to swarms of many small events and protracted volcanic tremor. The spectral range of volcanic signals is unequivocally broadband, with coincident high (>20 Hz) and very low (down to periods of hundreds of seconds) frequency signals frequently observed at many volcanic systems. As such, interpretations of volcano-seismic source and process require suitable characterisation in the time-frequency (T-F) domain. The adoption of automated approaches to routine seismic processing at volcanoes also creates the need to evaluate how we suitably extract discriminatory features of interest from such diverse volcano-seismic signals. Here we assess the performance of the continuous wavelet transform (CWT) for spectral representations of volcano-seismic signals. The localisation property of wavelet transforms gives the CWT a distinct advantage over commonly used moving-window Fourier transforms, enabling it to capture sharp changes in signal and represent signals over a wide range of timescales. Examination of seismic data for typical volcano-seismic phenomena, such as volcano-tectonic earthquakes (VTs), shows that CWT scalograms have better T-F resolution across broader frequency ranges than Fourier transform spectrograms, which suffer from greater spectral smearing in the time domain at higher frequencies. The inherent log-scale representation of CWT scalograms is also better suited for detection and representation of very-long-period (VLP) signals and for distinguishing volcanic signals from ambient microseismic noise. When applied to seismic data from Santiaguito volcano in Guatemala, CWT analysis reveals pre- and syn-eruptive signals across a wide range of frequency bands, ranging from 600 s to 50 Hz, with ultra-long-period signals (ULPs; 30 to 600 s) detected on instruments up to 1.9 km from the active vent, which is beyond the range of previously detected ULPs at this volcanic system. The CWT scalogram conveniently represents these simultaneous syn-eruptive spectral features in a single plot, which can aid exploratory analysis and inform source models. Furthermore, the ‘edge detection’ capabilities of the CWT accurately identify sharp changes in the raw signal over the VLP-ULP frequency range (5 to 600+ s), thought to represent sudden deflation associated with eruption, providing a useful tool for ‘picking’ explosive eruptions. The addition of an average wavelet energy distribution to CWT scalograms, which reveals the average energy across the whole signal at each wavelet scale, is also useful for characterising spectral content and identifying spectral peaks, as its smooth appearance is easier to interpret than FFT spectral amplitude plots. We conclude that wavelet transform methods are underutilised in volcano seismology, where their T-F localisation properties would be particularly well-suited, and suggest potential applications in terms of automated event detection and classification.