This paper proposes a method for estimating a set of spectral-power distributions of an omnidirectional light distribution from images taken by a camera aimed at a mirrored ball. First, we introduce some measuring systems that are designed for acquiring ambient light at any location in natural scenes. Second, a mapping algorithm is determined between coordinates on the ball and light rays in the world. Then we create images representing the omnidirectional light distribution. Third, the illuminant spectral-power distribution is recovered using the RGB image data and the calibrated reflectance data of the ball. Fourth, a CG image is created using the estimated omnidirectional illuminant distribution and a spectral-reflectance model for various objects. An experiment is conducted in our university gym.