Different approaches to measuring stimulus generalization in drug discrimination experiments are compared. The main issue, with implications for interpreting partial generalization, has been whether it is preferable to use graded or quantal indices. The first step is to determine whether discriminative drug effects are inherently quantal or graded; it would be pointless to attempt to assess a quantal phenomenon with a graded index. Evidence is reviewed suggesting that behavioural variables, and possibly pharmacological variables as well, determine the nature of the effects. Certain experiments used to evaluate the nature of generalization with exteroceptive stimuli have yet to be applied to the interoceptive stimuli that drugs produce. However, like many other attributes of drug action, discriminative stimulus effects are not immutable in nature, but are malleable and influenced by variables such as the schedule of reinforcement. There appears to be a potential for selecting procedures that generate graded responding in order to maximise precision. This ideal situation has yet to be fully realised in practice since the optimal conditions for generating graded responding remain to be identified, and present indices, based on percentage scores, do not generate interval scales of measurement. Until these issues are resolved, there is no case for maintaining that measurements of discriminative drug effects should with advantage be always quantal or always graded; either approach is compatible with research of high quality and can generate valid results and conclusions.