We describe a computerized method to measure the geometry of nanometer-scale data marks from AFM images. By compiling measurements of hundreds offeatures, we obtain statistically robust results, not only for mean values ofstructural parameters, but also for the standard deviations, so that process windows can be determined. On DVDs, we measured the following parameters: track pitch, bump height, bump width and length (at various threshold levels), bump length, and four sidewall slope angles, in each case reporting mean, standard deviation and other statistics. For each 10x10 pm image of a DVD stamper, containing about 100 bumps, we tabulated over 1000 values. In a plot of bump width vs. bump length, we found that width at half height increased from 328 nm for the shortest bumps (440 nm long) to about 385 nm for bumps longer than 800 nm; this matches the increase seen for corresponding optical signals produced when a finished disc is played. Where sidewall angle deviated from the norm, we were able to review the image data to identify the specific nature of the defect. Thus, feature geometry will no longer be a hidden variable in the path between controlling production equipment and observing the good or bad electrical performance of a finished disc.