1. A total of 312 cutaneous afferent units identified in the rat foot as belonging to one of nine major types of sensory receptors were included in the present study. A natural stimulus set was defined to differentiate optimally among those receptor types according to the distinguishing response patterns that it produced. It included air puffs, 30- and 300-Hz sinusoids, 200-mN force indentation of the skin, 1.2- and 6-N compressions of a skin fold, cooling the skin by 5 and 20 degrees C, warming by 5 degrees C, and heating by 15 degrees C. 2. The responses to predefined stimuli of 188 units were subjected to multivariate statistical analyses. The responses of an individual unit were measured as the number of impulses evoked by 10 stimuli, each lasting 10 s. Additionally, the number of impulses occurring for 5 s after withdrawal of a 200-mN indentation (1 of the 10 stimuli) was counted. 3. In discriminant analysis, the 11 stimulus variables predicted fairly correctly the grouping of afferent units into nine predetermined receptor categories (175 of 188, 93.1%), indicating a powerful ability to discriminate among different receptor types. Using hierarchical cluster analysis, afferent unit data described by 11 variables were divided into clusters that well represented prior receptor categories (170 of 188, 90.4%), suggesting the reliable application of this procedure to the classification of newly recorded cutaneous sensory receptors. 4. Eleven variables were then reduced to 7 on the basis of the results of factor analysis (95% of variance accounted for). The seven variables corresponded to 1.2-N compression, heating the skin by 15 degrees C, cooling the skin by 20 degrees C, 30- and 300-Hz sinusoids, withdrawal of a 200-mN indentation, and air puffs. 5. The seven selected variables correctly assigned afferent units into five modality-based categories in the discriminant solution (177 of 188, 94.1%). In the cluster solution, afferent units described by the seven selected variables were divided into clusters, most of whose members were modality specific (176 of 188, 93.6%). 6. The results indicate that cutaneous receptors can be divided into modality-specific groups according to similarities in their responses to seven stimulus variables. It is proposed that the stimulus set developed here and multivariate statistical methods can be used as powerful tools for the functional classification of central somatosensory neurons.