This study examines the four-dimensional (4D) seismic signatures from multiple seismic surveys shot during gas exsolution and dissolution in a producing hydrocarbon reservoir, and focuses in particular on what reservoir information may be extracted from their analysis. To aid in this process, hydrocarbon gas properties and behaviour are studied, and their relationship to the fluid-flow physics is understood using numerical simulation. This knowledge is then applied to interpret the seismic response of a turbidite field in the UK Continental Shelf (UKCS). It is concluded that for a repeat seismic survey shot 6 months or more after a pressure change above or below bubble point (as in our field case), the gas-saturation distribution during either exsolution or dissolution exists in two fixed saturation conditions defined by the critical and the maximum possible gas saturation. Awareness of this condition facilitates an interpretation of the data from our field example, which has surveys repeated at intervals of 12–24 months, to obtain an estimate of the critical gas saturation of between 0.6 and 4.0%. These low values are consistent with a range of measurements from laboratory and numerical studies in the open literature. Our critical gas-saturation estimate is also in qualitative agreement with the solution gas–oil ratios estimated in a material balance exercise using our data. It is not found possible to quantify the maximum gas saturation using the 4D seismic data alone, despite the advantage of having multiple surveys, owing to the insensitivity of the seismic amplitudes to the magnitude of this gas saturation. Assessment of the residual gas saturation left behind after secondary gas-cap contraction during the dissolution phase suggests that small values of less than a few per cent may be appropriate. The results are masked to some extent by an underlying water flood. It is believed that the methodology and approach used in this study may be readily generalized to other moderate- to high-permeability oil reservoirs, and used as input in simulation model updating.