In Reply.
—We thank Dr Lempert for raising an interesting issue by his sophisticated analysis of our technical description of computerized fundus analysis.Dr Lempert is indeed correct in calculating that with the Humphrey Retinal Analyzer (Allergan Humphrey, San Le andro, Calif), a 1-pixel horizontal retinal disparity in the stereo pair indicates a depth of about 100 μm. However, in stating that this factor by itself can account for the variability calculated by us, 95% confidence intervals for individual depth measurements ranging from 118 to 392 μm, Dr Lempert has probably assumed that the retinal disparity can be detected only in integral pixel values, that is, 1, 2, and 3, etc. This assumption is not correct.The software of this analyzer can detect the horizontal retinal disparity in fractions of pixel values in the following way. The "scene matching" of the stereo pair proceeds in stages from a coarse level