A long literature provides empirical estimates of differential treatment (by race and other factors) in the criminal justice system, taking sentencing rules as given. This article considers the disparate impact of the rules, rather than their implementation, by providing a quantitative assessment of the Maryland sentencing grid for person offenses using the distribution of defendants on the grid. The basic insight is that movement along the grid itself (either across rows or across columns) is nonlinear in sentence lengths (in both absolute magnitudes and percent terms). As a result, an additional point results in a different increase in the recommended sentence length depending on the other factors that are present. Because the distribution of factors is not random, factors like weapon use that appear to have equal weight in the formula developed by policymakers turn out to have very different weights in practice.