FIG. 7. Testing selectivity of C2 units to bars and gratings. We measured the responses from a population of C2 units fit to the population of V4 neurons from Pasupathy and Connor (2001), using bars and gratings. Three summary plots are presented: *A*: orientation bandwidth; *B*: bimodal tuning index; and *C*: tuning to Cartesian, polar, and hyperbolic gratings. *A* shows a histogram of orientation bandwidths measured for 97 of the C2 units that showed significant response to bar stimuli. The median orientation bandwidth for the C2 population was 51.7°. *B* shows a histogram over bimodal tuning index for the same 97 model C2 units. The median bimodal indexes over the population of C2 units is 0.12. *C* shows a summary plot of all 109 model C2 units to Cartesian, polar, and hyperbolic gratings. The grating stimuli, analysis procedure, and plotting convention used in Gallant et al. (1996) are reproduced to asses the selectivity to complex gratings. For each C2 unit, the maximum responses to each grating class (Cartesian, polar, and hyperbolic) form a 3-dimensional vector, normalized to unit length, and plotted in a 3-dimensional space with each axis representing the response to a grating class (the viewpoint is oriented so that the origin of the coordinate system is at the center, and the vector whose responses are equal is pointing directly out of the page). The vector for each model C2 unit is plotted as a circle with the size of the circle indicating the magnitude of the highest response over all of the grating stimuli. The bias toward polar and hyperbolic gratings, which is a characteristic previously described in V4 (Gallant et al. 1996), indicates that for most C2 units, the optimal stimulus was non-Cartesian. Our results show a stronger bias than reported in (Gallant et al. 1996).