A method is described to characterizing the digital camera.The nonlinear relationship between the RGB signals generated by digital camera and original image CIE XYZ values is obtained using BP neural networks.The reasonable structures（3—10—10—3）of the BP neural networks are found by researching on NIKON D200 cameras

yielding a modeling accuracy typically average color difference 1.9～2.2 ΔE units and maximal color difference 6.7～7.4 ΔE units.The repeatability of the experimental setup is studied and the influence of the quantity of the samples is also discussed.The experiments results show that the BP neural networks can be used to characterize commonly digital camera.