A model to predict voluntary water intake (WI) of a pig fed a known diet in a known environment is described. The daily retentions of protein, lipid, water and ash were estimated over time using a published pig growth model. Food intakes were estimated using published methods. WI was estimated by adding the amounts required for digestion (WD), faecal excretion (Wfec), growth (WG), evaporation (WE), urinary excretion (WU) and by then subtracting the water arising from feed (WF), from nutrient oxidation (WO) and synthesis of body constituents (WS). WD was predicted assuming an absorption of water of 0·10, 0·16 and 0·07 kg/kg digestible carbohydrate, crude protein and lipid respectively. Wfec was estimated taking into account the water associated with the undigested protein (0·86 kg/kg), diethyl ether extract (-12·11 kg/kg), crude fibre (1·86 kg/kg), ash (-0·42 kg/kg) and N-free extract (4·4 kg/kg). The basal level of WE was estimated from the heat production of the pig fed ad libitum (MJ/d) as: 0·25×(metabolizable energy-energy retained as protein and lipid)×0·4, where 0·25 is the assumed proportion of the insensible heat loss at the comfort temperature and 0·4 is the water lost per MJ dissipated heat. WE in a hot environment was predicted by assuming that evaporation increased up to three times the basal level to offset the decreased sensible heat loss. To predict WU a water requirement for renal excretion of 2·05 and 3·40 kg/osmol excreted N as urea and minerals respectively was assumed. The urinary load of N and minerals was predicted from the intake of digestible nutrients and their retention. From the oxidation of 1 kg carbohydrate, protein, and fat it was assumed that 0·6, 0·42 and 1·07 kg water (WO) were released respectively. WS was predicted by assuming a release of 0·16, 0·07 and 0·57 kg water per kg retained protein, retained lipid coming from digestible lipid, and retained lipid coming from digestible carbohydrate respectively. The model is strongly rooted in a theoretical structure. When its predictions were compared with data from suitable experiments, the results were not significantly different. Both the pattern and the magnitude of responses of the model to changes in body weight, feed intake and environmental temperature are sensible and it allows a fuller prediction of voluntary water intake than the methods currently available.