Bullet matching is a process used to determine whether two bullets may have been fired from the same gun barrel. Historically, this has been a manual process performed by trained forensic examiners. Recent work, however, has shown that it is possible to add statistical validity and objectivity to the procedure. In this article, we build upon the algorithms explored in Automatic Matching of Bullet Lands (Hare, Hofmann & Carriquiry (2017), Automatic matching of bullet lands. ArXiv E-Prints) by formalizing and defining a set of features, computed on pairs of bullet lands, which can be used in machine learning models to assess the probability of a match. We then use these features to perform an analysis of the two Hamby (Hamby, Brundage & Thorpe (2009), The identification of bullets fired from 10 consecutively rifled 9 mm Ruger pistol barrels: a research project involving 507 participants from 20 countries. AFTE J., 41, 99–110) bullet sets (Set 252 and Set 44), to assess the presence of microscope operator effects in scanning. We also take some first steps to address the issue of degraded bullet lands and provide a range of degradation at which the matching algorithm still performs well. Finally, we discuss generalizing land-to-land comparisons to full bullet comparisons as would be used for this procedure in a criminal justice situation