This study examines whether criminal suspects facing more severe punishments are more likely to go to trial. Sample selection makes it difficult to obtain valid proxies for severity; for instance, I expect severity to be positively related to the prosecutor’s decision to indict, to indict in federal court (versus state court), and to try the suspect. Theoretical and empirical findings indicate that in samples containing only indicted, convicted, or tried suspects, reasonable proxies for severity may be negatively related to actual severity. The assignment of defendants to judges randomizes the severity of punishment in a manner that is unrelated to sample selection. Thus, by examining the effect of these assignments, I find that a 10-month increase in prison sentences raises trial rates by 1 percentage point.