Perceived contrast of narrowband spatial patterns is directly relatable to contrast sensitivity and discrimination (Cannon & Fullenkamp 1991, 1995), but it is unknown whether band-limited contrast sensitivity can predict the perceived contrast of broadband, complex imagery like natural scenes. On the other hand, the phenomenon of contrast constancy (Georgeson and Sullivan, 1975) has been invoked to suggest that the visual system equalizes contrast responses across scale/spatial frequencies. To resolve this longstanding question, we measured perceived contrast weights across spatial frequency during viewing of photographic scenes. On each trial an image, a photograph of a real-world scene, is divided into two sets of eight 1-octave frequency bands, each of which has its amplitude varied by a random amount, and the two sets are reassembled into two test images. The observer chooses which of the two appears to have higher contrast, i.e. a larger range of grayscale values. Weighting functions can be derived as the correlation between trial-to-trial band weights and observer choice (chosen or rejected). We find that the weighting function peaks at mid-high spatial frequencies, with considerable variation between subjects, indicating that observers are not using a flat broadband contrast statistic (e.g. RMS contrast) to make their choices. Weighting function shapes are found to be dependent on image statistics including edge density and spatial frequency amplitude spectrum slope. Simulations equating perceived contrast with contrast response, incorporating predictions based on the effects of contrast gain control, produce similar results as the human observers.