The Permian–Triassic boundary interval in shallow shelf seas of South China shows Upper Permian limestones overlain by lowermost Triassic microbialites. Global sea-level rose across the Permian–Triassic boundary, but an irregular top-Permian erosion surface across a 10 km north–south transect of the Great Bank of Guizhou contains evidence of sea-level fluctuation. The surface represents the ‘event horizon’ of mass extinction, below the biostratigraphic Permian–Triassic boundary defined by first appearance datum of conodont Hindeodus parvus. An Upper Permian foraminiferal grainstone beneath this surface contains geopetal sediments, etched grains, and pendent and meniscus cements interpreted here as vadose. However, these latter diagenetic processes occurred before the event horizon and were followed by erosion of the final Permian surface. This erosion cuts previous fabrics but lacks evidence of weathering or bioerosion. A few centimetres below is an earlier grainstone that was also eroded but lacks proof of sub-aerial processes. Samples therefore reveal one, or possibly two, small-scale relative sea-level changes before the Triassic transgression in this area, and these may relate to local tectonics. The final Permian surface is subject to at least four interpretations: (i) sub-aerial physical erosion and dissolution by carbon dioxide-enriched fresh water or carbon dioxide-enriched mixed water, prior to Triassic transgression; (ii) sub-aerial physical erosion overprinted by dissolution related to carbon dioxide-enriched sea water in the Early Triassic transgression; (iii) submarine dissolution affected by acidified sea water due to rapid increase in volcanically-derived carbon dioxide and oxidized methane released from marine clathrates; (iv) submarine dissolution due to acid anoxic waters rising across the continental shelf, unrelated to atmospheric carbon dioxide or oxidized methane. Field and petrographic evidence suggests that (i) is the simplest option; and it is possible that (ii) and (iii) occurred, but none are proved. Option (iv) is unlikely given the evidence and modelling of supersaturation of upwelled waters with respect to bicarbonate.