This paper presents a new evaluation method of the color estimation accuracy for a set of multispectral color sensors. In this method, the probability density distribution of the spectral reflectance is assumed, and then it yields the possible range of estimation error in the color space at a specified confidence level. It is shown that the error range in the color space is expressed as the ellipsoids, when the probability density of the spectral reflectance is given by a multi-dimensional Gaussian function. Some examples of the evaluations of multispectral imaging systems are shown for different numbers of bands, and the results are compared with MacAdam ellipse.