Sunstein et al. (2006) utilized the random assignment of judges to cases in the U.S. Courts of Appeals to estimate the effect of partisanship on these judges without the possibility of bias from unobserved heterogeneity. This article critiques and improves their study by controlling for changes in circuit composition over time, including cases in issue areas that Sunstein et al. omitted, and omitting cases from circuits where judges were not randomly assigned. I find that Sunstein et al. slightly underestimated the effect of partisanship in the courts of appeals, failed to find evidence of partisan effects for issue areas in which judges are affected by partisanship, mistakenly found evidence of partisan effects for issue areas in which judges are not affected by partisanship, and underestimated the degree to which partisanship varies between circuits. Random judicial assignment offers promising possibilities for the study of judicial decision making, but care must be taken in order to reap the benefits of these natural experiments.