Drug discrimination has been an important technique in behavioural pharmacology for at least 40 years. The characteristics of drug-produced discriminative stimuli are influenced by behavioural and pharmacological variables, including the doses used to establish discriminations. This review covers studies on the effects of varying the training dose of a drug in a search for general principles that are applicable across different drug classes and methodological approaches. With respect to quantitative changes, relationships between training dose and the rate of acquisition or magnitude of stimulus control were found for most drug classes. Acquisition accelerated with dose up to a point beyond which drug-induced impairments of performance had a deleterious impact. Sensitivity to the training drug as measured by ED(50) values typically increased when the training dose was reduced. Qualitative changes were more complex and appeared to fall into three categories: (a) changes in profiles of generalization between partial and full agonists; (b) reduced specificity of some discriminations at small training doses; and (c) changes in the relative salience of actions mediated through different neurotransmitter systems or from central and peripheral sites. Three-lever discrimination procedures incorporating 'drug versus drug' or 'dose versus dose' contingencies enabled detection of more subtle differences than the simple 'drug versus no drug' approach when applied to the opioid, hallucinogen and barbiturate classes of drugs. These conclusions have implications for the interpretation of data from studies that use either within-subject or between-subject designs for studying the discriminative stimulus effects of drugs.