Thirty-six diets were formulated from a pool of 33 ingredients (plus a mineral and vitamin supplement) to give the following composition: crude protein 140 or 200 g kg-1; oil 20, 40 or 80 g kg-1; crude fibre 25, 50 or 100 g kg-1. The digestible energy (DE) and metabolisable energy (ME) contents of the diets were determined with 4 pigs (approx. 50 kg liveweight) per diet. The diets were analysed in several laboratories for a wide range of chemical components (3-5 laboratories per component). The DE of the diets measured in vivo agreed well with the DE calculated from assumed values for the ingredients (DE calculated - DE in vivo = -0.22 MJ; standard deviation = 0.461 MJ). Neutral detergent fibre (NDF) had the highest correlation coefficient with DE (-0.907), and in 2- or 3-factor equations derived to predict DE from chemical composition NDF and gross energy (GE) were particularly effective. The best measure of the effect of fibre content of the diets on energy value was always NDF. The most effective 4-factor linear equation was DE(MJ kg-1 DM) +17.49-14.9 NDF+15.7 OIL(PT)+7.8 CP-32.5 ASH. and the best equation with interaction terms was DE (MJ) kg-1 DM)=17.05-11.9 NDF+123 OPT squ+28CP.NDF-116 ASH.NDF. where OIL (PT) = petroleum ether extract; OPT squ = petroleum ether extract squared; CP = crude protein, and components are in kg kg-1 DM. After allowing for laboratory and animal variation, the linear equation would be expected to give 79% of predictions within ±0.5 MJ kg-1 DM of the true value. Replacing NDF with crude fibre reduced this to 60%.