Hypothetical neural codes underlying the sensation of tactile roughness were investigated in a combined psychophysical and neurophysiological study. The stimulus set consisted of plastic surfaces embossed with dot arrays of varying dot diameter and center-to-center spacing. Human subjects explored each surface with the pad of the index finger and reported their subjective sense of roughness magnitude. The same surfaces were scanned across the receptive fields of cutaneous mechanoreceptive afferents in monkeys while recording the evoked action potentials. Hypothetical neural codes for roughness magnitude were computed from the neural response patterns and tested for their ability to account for the psychophysical data. The psychophysical results showed that subjective roughness magnitude is an inverted U-shaped function of dot spacing that peaks near 3.0 mm spacing, and that increased dot diameter produces decreased roughness sensations at all dot spacings. Hypothetical neural codes that do not bear a consistent relationship to roughness magnitude across all of these stimulus conditions can be rejected as the code for roughness. Four types of neural codes were considered. They were based on (1) mean firing rate, (2) general variation in firing rate, (3) short-term temporal variation in firing rate, and (4) local spatial variation in firing rate. Mean firing rate failed to explain the psychophysical results: surfaces that evoked the same firing rate often evoked very different roughness judgments. In contrast, neural codes based on firing-rate variation, especially in slowly adapting afferents, account for the psychophysical results.