An experiment was conducted in which a large group of subjects was asked to categorize speech stimuli. In each subexperiment, two groups of 10 subjects categorized, with immediate feedback, 90 stimuli consisting of six monosyllables arranged in 15 randomized blocks. One group, the feature-contrast group, could solve the categorization task on the basis of a feature contrast or rote memory. The second group, a control group, could operate only on the basis of rote memory. Data are presented for the following features: ± vocalic, ± voice, ± nasal, ± continuant, and ± strident. Results indicate that the nasal, strident, and vocalic features have conceptual reality, that pairing a conceptually real feature with a nonoonceptually real feature does not improve performance, and that the data are not easily related to many existing notions or data concerning distinctive-feature theory.