Reef islands on the Great Barrier Reef are influenced by a range of environmental factors. A meta-analysis of 103 islands is presented to express variation in island size (area and volume) as a function of latitudinal and cross shelf gradients in regional oceanographic factors (exposure to incident waves, tidal range and tropical cyclone frequency) and local physical factors (position on the shelf, area, length and depth of supporting reef platform, vegetative cover). Models performed well for unvegetated sandcays (R2 = 0.89), vegetated sandcays (R2 = 0·72) and low wooded islands (R2 = 0.78), with a moderate level of variation explained when all islands were simultaneously regressed (R2 = 0.58). Future island dynamics were simulated for anticipated changes in cyclone regime, wave activity and sea level. For 38 islands mapped on the 1973 Royal Society and Universities of Queensland Expedition to the Northern Great Barrier Reef, change over the same 22 year period (1973–1995) was determined and the relative magnitude of observed and modelled changes was compared and found to be consistent through rank correlation analysis (Γ = 0.84 for unvegetated sandcays, Γ = 0.81 for vegetated sandcays). Simulations of island area or volume change from 2000 to 2100 indicated that under a 30% decrease in tropical cyclone activity, unvegetated sandcays continue to accrete at a lower rate, whereas all island types erode under a 38% increase in tropical cyclone activity. Vegetated sandcays initially accrete at higher levels of cyclone activity, entering an erosive state with a 60% increase in activity. Low wooded islands are unresponsive to environmental changes modelled. A sensitivity analysis of vegetated and unvegetated sandcays indicated that the presence of vegetation increases the tropical cyclone activity threshold at which islands begin to erode. Greatest sedimentary losses occur within the central band of high cyclone activity between Cooktown and Mackay.