The homogeneity of the restraint construct is investigated in a study of 54,525 subjects participating in a weight reduction program. Using the German version of the TFEQ (Stunkard & Messick, 1985, Journal of Psychosomatic Research, 29, 71-83; Pudel & Westenhoefer 1989a, Fragebogen zum Essverhalten; Handanweisung), it is shown that different types of relations exist between the items measuring dietary restraint and the disinhibition scale. A discriminant analysis in a subgroup of moderately highly restrained eaters with either low or high disinhibition (n = 1759) reveals different sets of restraint behaviours and cognitions that differentiate between high and low disinhibition. These findings are replicated in an independent validation sample (n = 1693). Two restraint subscales are constructed, one associated with increasing disinhibition, the other with decreasing disinhibition. It is argued that different sets of restrained behaviours and cognitions should be differentiated when looking at the causal link between restraint and disturbances of eating behaviour.