The sensitivity of speech discrimination tests is defined as being large if a small change in experimental conditions causes a large and repeatable change in test score. The aim of this paper is to investigate ways by which test sensitivity can be maximized. The effect of test difficulty on sensitivity is examined by the use of a model based on signal detection theory and subjective similarity rating concepts. For several sets of assumptions about the parameters underlying the perceptual processes, the model indicates that test difficulty should be in the range 73% to 87% to achieve the maximum sensitivity with speech tests using between four and eight response foils. The model results also indicate that the sensitivity increases with the number of response foils. When the variation of test reliability with test difficulty is taken into account, the test difficulty which maximizes the sensitivity/reliability trade‐off lies in the range 85% to 90%. It is argued that irrespective of the optimum test difficulty, a maximally sensitive test will contain items equal to one another in difficulty.