In order to improve the accuracy of colorimetric characterization of digital image devices, an adaptively polynomial model based on Gaussian weighted function is proposed. Certain nearest training samples are adaptively selected for each target sample, and a Gaussian function is designed to weight the contribution of each training sample, and then the least-squares method is adopted to get the coefficients of the corresponding polynomial models. An experiment is implemented to colorimetric characterization of digital still cameras with the proposed model, common polynomial model and polynomial model based on inverse distance weighted function. The experimental results indicate that the prediction accuracy of the proposed model is highest in all the training and testing samples combinations.