Abstract: Stochastic photon noise is one of the most important factors which influence the quality of X-ray microscope images. The influence of the noise is described by the signal-to-noise ratio (SNR). Generally, increasing the number of photons yields better SNR, but inevitably with the side effect of a higher dose for the specimens. To find a good balance, one has to find a way to minimize the photon number while keeping an acceptable SNR, but conventional empirical Rose criterion is of little help, especially when applied to processed noised images. Here a concept of “probability-to-distinguish” was provided as an improvement on the Rose criterion. It can be used to find out the threshold of the photon number. The new method was applied to analyze noised images in image subtraction and compute the minimum photon number required. The conclusion is that objects can be distinguished from their background when the probability-to-distinguish is set at above 0.4.