Introduction:The human ventral cortical stream contains a mosaic of different areas responding selectively to different categories of visual stimuli. In particular,three occipito-temporal areas have been shown to be scene-selective: the parahippocampal place area (PPA), the retrosplenial cortex (RSC) and theoccipital place area (OPA). These regions are known to support high-level functions during scene perception such as navigation, scene recognition, and contextual associations. However, whether and how these regions extract more basic visual features in scenes in order to yield high-levelrepresentations remains unclear. In the present fMRI study, we specifically investigated whether the PPA, the RSC and the OPA extract spatialfrequencies in scenes. Low spatial frequencies (LSF) carry coarse information and are characterized by a high luminance contrast. Conversely, highspatial frequencies (HSF) are related to fine details and are characterized by a low luminance contrast. In order to avoid any confusion between effects of spatial frequencies and contrast, recent studies equalize contrast between LSF and HSF. In the present study, we also examined the effect of contrast equalization between spatial frequencies within these regions.Methods:Sixteen right-handed participants (9 males; 23 ± 2 years) performed two experiments: the Spatial Frequency (SF) and the Localizer Experiments. Inthe SF experiment, they performed a categorization task on grayscale photographs of scenes (indoor and outdoor with 24°x18° of visual angle),filtered in LSF (cutoff frequencies: 0.5, 1, 2 cpd), HSF (cutoff frequencies: 3, 6, 12 cpd) and non-filtered (NF). In a first condition, contrast was notequalized between the filtered and NF scenes. In a second condition, contrast was equalized between the filtered and NF scenes (Fig.1). A block design was used and consisted of eight functional runs (four per Contrast condition). Each functional run was composed of nine 25-sec task blocks (one block per spatial frequency cut-off and three blocks of NF scenes), including 10 scenes (5 indoors and 5 outdoors), interspersed with three 25-sec blocks with a fixation dot in the center of the screen displayed against a grey background. Each scene was displayed for 100 ms, and masked for 30 ms (1/f noise). In the Localizer Experiment, participant performed a one-back task on grayscale photographs of scenes, faces and objects. A block design was used, consisting of one run, composed of eight 15-second task blocks (2 blocks of faces, 2 of objects and 4 of scenes), including 15 different photographs of the same type, interspersed with four 15-second fixation blocks. For each participant, the PPA, the RSC and the OPA were identified in both hemispheres by a [scenes>objects+faces] contrast and ROIs were created using 3-mm spheres centered on individual peaks. Percentage signal change in the SF Experiment was then extracted from these ROIs in each participant and submitted to a repeated-measure ANOVA.Results:The PPA was more strongly activated by LSF and NF scenes than by HSF scenes when contrast was not modified, but was selective to HSF whencontrast was equalized across spatial frequencies suggesting that PPA activity relies on an interaction between spatial frequency and contrast in scenes. In the RSC, LSF and NF scenes elicited stronger response than HSF scenes when contrast was not modified, but there was no effect of spatial frequencies when contrast was equalized across the filtered scenes suggesting that the RSC is sensitive to high contrast differences. Finally, the OPA was selective to HSF scenes, irrespective of the contrast condition (Fig. 2).Conclusions:The present results demonstrate that scene-selective regions extract low-level properties of scenes such as spatial frequencies and contrast, withdistinct response pattern in each region suggesting different functional properties.