In this paper, a method for generating reflectance spectra from camera signals is proposed. It consists of two main steps. The first step is to characterize the camera by estimating the mapping matrix from colour signal directly to reflectance using a training data set including camera signals and their reflectance functions. The second step is to construct a new reflectance function based on the one generated in the first step via a constrained least squares method. The great advantage of this method over conventional methods is without knowing the information of camera sensors and illuminant. The method together with the conventional methods such as those developed by Wiener, and Hasegawa and Fairchild were tested using two types of data sets: the simulated and real camera data. The results showed that the new method outperformed the conventional methods, especially for the real camera data