A method is described for estimating the illuminant spectra of an omnidirectional light distribution from the images of a camera aiming at a mirrored ball. First, we introduce an imaging system using a spherical steel ball and a normal color CCD camera. The parameters including the optical axis and the focal length of the camera lens are estimated to determine the mapping between coordinates on the ball and light rays in the world. We create images to representthe omnidirectionallight distributionin the world. Next, algorithms are presented for estimating the illuminant spectra from the image data. In experiments, the spectral radiance distributions are estimated in an outdoor scene including strong direct sunlight. Finally, a CG image is demonstratedusing the estimated illuminant distribution.