In this paper, we propose a system for the identification of the system that has produced a Digital Radiography (DR) image. It takes advantage of the statistical properties of the noise left by any DR systems. In particular, a three parameter exponential model of the relationship in-between the image intensity and the overall noise variance is suggested. Its parameters are used as input of a classifier learned in order to discriminate different DR systems. Experiments conducted on images issued from 5 different DR systems show it is possible to identify with good accuracy the origin of one DR image.