We have examined the reliability of several food energy assessment systems for healthy men. The predictions of metabolizable energies were compared with determinations made in energy balance studies with three maintenance diets (12 MJ/d); one of the diets was moderate in non-starch polysaccharide (NSP; 2.1% of gross energy) and two were higher in NSP (3.5-4.6% of gross energy). A fourth diet was a submaintenance (6 MJ/d) high NSP (7% gross energy) diet. Discrepancies between the different food energy assessment systems and the determined metabolizable energy values ranged between 0 and 15%. With the maintenance diets, the Atwater specific factor system had errors generally within 6% of the determined value and a limit of agreement (bias +2SD) for diets of 10%. This accuracy compares with errors of 2% for both the originally published assessment of this system and a more recent general formula; both systems were without bias with increasing NSP content of the diets but the latter had limits of agreement within 3%. In contrast, the Atwater general, the European general and a recent FDA general formula showed increasing bias with increasing NSP intake. All of the general energy assessment systems overpredicted metabolizable energy from the high NSP submaintenance diet, which shows that even the least biased and most reliable energy assessment system that we identified applies to maintenance diets only; thus a correction has to be made for submaintenance diets.