A masking paradigm was employed to measure the spatial frequency selectivity of channels underlying human stereopsis. Observers viewed spatially filtered (0.4 octave bandwidth) random-dot stereograms in which a disparate bar appeared in either the top or bottom half of the display; superimposed on one RDS half-image was a noise target whose spatial frequency content was varied relative to that of the RDS. A staircase procedure was used to measure the monocular noise energy (and hence the signal-to-noise ratio) at which observers could judge the location of the disparate bar on 71% of trials. Statistical analyses showed that the resulting stereoscopic masking functions could be grouped into two sets, one with peak sensitivity at 3 c/deg and the other with peak sensitivity at 5 c/deg. These two channels were observed for both crossed and uncrossed disparities ranging from coarse to fine. Essentially the same results were obtained with binocular noise and with stereo displays flashed too briefly to be affected by eye movements. Our results are inconsistent with models of stereopsis in which the disparity range to which a channel is sensitive varies with that channel's peak spatial frequency. These data imply that the spatial frequency selectivity of stereopsis differs from the tuning of spatial channels underlying the detection and discrimination of form.