Multiple factors have affected the decline in autopsy rates. Our goal was to determine the relation of physicians' recommendations regarding autopsy, as well as patient and surrogate decision-maker characteristics, to autopsy performance. We assessed measures related to autopsy performance using data from two teaching institutions in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments. We included patients who had died within 6 months of their index hospitalization and for whom information was available on autopsy performance, physicians' response to questions about autopsy, and interviews with surrogate decision makers about autopsy performance. We assessed the association between autopsy performance and the strength of a physician's recommendation for autopsy, adjusting for patient, surrogate, and physician characteristics. Of the 680 patients who died, 59% (n = 402) met our inclusion criteria. Based on physician and surrogate responses, the expected autopsy rate was 42% while the actual autopsy rate was 23%. The autopsy rate was higher when the physician's recommendation for autopsy was strong or very strong at the time of death compared with when autopsy was not recommended strongly or not at all (P <0.001). The strength of the physician's postmortem recommendation was independently associated with autopsy performance after adjusting for patient, surrogate, and physician characteristics (P <0.001). Autopsies are less likely to be performed when not recommended strongly or not at all. Training physicians (or others) how to recommend autopsies may increase autopsy rates.