Standard models of texture segregation comprise (i) a linear filter (to enhance a constituent texture), (ii) a nonlinearity (to compute 'texture energy'), (iii) a second linear filter (to enhance the texture modulation), and (iv) a decision stage. We estimated the orientation tuning of the 2nd-order filter using the critical-band masking technique. Methods: The task was 2-interval forced-choice detection. Target stimuli were 2nd-order vertical sine-wave gratings (0.5 cycle/deg) masked by 2nd-order noise. The noise had narrow spatial frequency bandwidth centered on the target frequency. The orientation bandwidth of the noise was varied across blocks, ranging from 0 (no noise) to 180 deg (all-pass in the orientation domain) in steps of 30 deg, by sweeping the cutoff angle from horizontal (at 0 deg bandwidth) through vertical (at 90) and back to horizontal (at 180 deg), both clockwise and counterclockwise. In the target interval, a grating plus noise mask was used to modulate between two carrier textures (sine-wave gratings, 4.0 cycle/deg, oriented at -45 and 45 deg). In the non-target interval, the modulator was a noise mask alone. Modulator and carrier grating phases were randomized. Noise spectral density was held constant, and signal contrast was controlled by interleaved staircases. Threshold contrast elevation was obtained for each noise-mask cutoff angle. A sigmoid was fit to clockwise and counterclockwise noise-orientation-cutoff sweeps to derive the 2nd-order channel's orientation tuning. Results: Estimated 2nd-order orientation tuning is broad relative to 1st-order channels. Simulations of the standard filter-rectify-filter texture-segregation model will be compared to our behavioral data.