The diagnostic process in sialography was evaluated using the mouse parotid gland as a model. The relative significance of each of the three diagnostic phases, psychophysical, psychological and nosological, was determined by analysing observer accuracy in detecting sialoadenitis and sialodochitis on digital and analogue microsialograms. Automated diagnosis based on parametric values of the digitized image was compared with the human system using ROC analysis. The overall accuracy of automated diagnosis was superior to that of an observer with analogue images. It is concluded that if observers were to be given the same quantitative information and informed of its diagnostic value in the psychological phase, then variation in the detection of the abnormalities could be reduced and the overall diagnostic accuracy in sialography improved.