A novel approach to patient dose and image quality optimization was developed and implemented for chest and lumbar spine radiography. A Monte Carlo model of the imaging chain, including an anthropomorphic voxel-phantom to simulate the patient, was utilized. Detector noise and system unsharpness were modeled and their influence on image quality considered. Image quality was quantified by the contrast ((Delta) OD) and the ideal observer signal-to-noise (SNR) for a number of relevant image details at various positions in the anatomy and measures of dynamic range (DR). Among systems evaluated in a clinical trial, a reference system, acknowledged to yield acceptable image quality, was selected. A large variety of other imaging conditions were simulated and compared to the reference system. Some of the simulated systems were found to give as good imaging performance but at substantially reduced patient doses: 35% and 50% reduction in the lumbar spine AP and the chest PA view, respectively. The model was also used to define a single-valued 'figure-of- merit,' the physical image quality score, PIQS, with the aim to make possible ranking of the imaging systems. By comparing the ranking according to PIQS with radiologists' ranking it was possible to analyze the features in the images which are clinically important.