We use practice variation across physicians to uncover the role of medical care in causing opioid dependence. Using health records of 2 million US veterans with emergency department visits, we find that quasi-random assignment to a top (versus bottom) decile prescribing provider significantly increases subsequent opioid use and misuse rates. Instrumental variable results show that opioid prescription receipt leads to a 20 percent increase in the probability of long-term prescription opioid use and sizable increases in the development of opioid use disorder and opioid overdose mortality. We find suggestive evidence of transition into illicit opioids due to prescription opioid exposure.