Image quality captured with five different spectral sensitivities were examined. Images were created with an image sensor model using four spectral images and three exposure index values. A psychophysical experiment was conducted and interval scales of preference were derived. It was shown that, as the exposure index increased, the preferred image sensor parameters shifted to better noise but less accurate color reproduction though strong image dependence was also shown. The interval scales were modeled using color difference, noise amplitude and their covariance. According to the fitting result, it was shown that even though both color difference and noise were measured in the same units, the effect of noise on the image quality was about 40% or less unless the high level perception of human visual system came into play.