The extent to which rules set by the legislature bind or influence decisions regarding sentence length is central to institutional design and to determining the practical impact of any proposed reform regarding criminal punishment. However, it is generally difficult to identify empirically the impact of sentencing recommendations because court actors may have preferences that are correlated with those outlined in the guidelines. In this article, we take advantage of a new source of identification to study how government actors interact and make decisions in the criminal sentencing process. We identify instances in the Maryland circuit court in which the case facts are not consistent with the final sentence recommendation—inconsistencies that appear to be the result of human error and exogenous to the preferences of downstream actors. We find that even an advisory guidelines system like the one in Maryland has a direct impact on judicial decision making in cases involving drugs and violent crimes. Judges appear eager to go along with an erroneous lesser sentence for violent offenses. In contrast, judges appear to discount mistakes that are too high. This asymmetry does not occur for property and drug offenses that are simpler and more frequently encountered. More generally, experience matters. Error rates are lower for more frequently occurring offense types and lower for those court professionals who complete more of the sentencing worksheets. The net effect of sentencing guidelines on time served appears to be small because parole boards counteract the remaining influence of the guidelines.