This paper analyzes what portion of US nonfatal crashes are alcohol-involved and how well police and hospitals detect involvement. A capture recapture model estimated alcohol involvement from levels detected by police and hospitals and the extent of detection overlap. We analyzed 550,933 Crash Outcome Data Evaluation System driver records from 2006-2008 police crash report censuses probabilistically linked to hospital inpatient and emergency department (ED) discharge censuses for CT, KY (admissions only), MD, NE, NY, SC, and UT. We computed national estimates from NHTSA's General Estimates System.Nationally an estimated 7.5% of drivers in nonfatal crashes and 12.9% of nonfatal crashes were alcohol-involved. (Crashes often involve multiple drivers but rarely are two alcohol-involved.) Police correctly identified an estimated 32% of alcohol-involved drivers in non-fatal crashes including 48% in injury crashes. Excluding KY, police in the six states reported 47% of alcohol involvement for cases treated in EDs and released and 39% for admitted cases. In contrast, hospitals reported 28% of involvement for ED cases and 51% for admitted cases. Underreporting varied widely between states. Police reported alcohol involvement for 44% of those who hospitals reported were alcohol-involved, while hospitals reported alcohol involvement for 33% of those who police reported were alcohol-involved. Police alcohol reporting completeness rose with police-reported driver injury severity. At least one system reported 62% of alcohol involvement. Police and hospitals need to communicate better about alcohol involvement. Despite the proven effectiveness of brief alcohol intervention, EDs rarely detect, much less intervene with crash-involved drinking drivers. Both police and EDs particularly need to assess alcohol involvement in minor injury better.