This is an ethnographic account of peer learning among a group of heroin 'addicts' in Delhi, India. This study responds to the limited attention given to 'naturalistic' or 'informal' peer learning patterns in the educational literature, and the lack of explicit exploration of the phenomenon among drug user populations. The study involved seven and a half months of fieldwork with the predominant use of participant observation and semi-structured interviews to generate data. Analysis was inductive and interpretive with the use of situated learning theory to 'tease out' patterns in the data. The participants were using and non-using addicts affiliated to SHARAN, a non-governmental organization (NGO) in the religious marketplace of Yamuna Bazaar. The group included approximately 300-500 members, 20 of whom were main informants. Analysis of the group organization revealed community-based and masculinity-based characteristics that enabled the group to manage stigma, promote 'positive' ideals, and co-construct nonhegemonic masculinities. Peer-based outreach was identified as a form of 'institutional' peer learning in which peer educators performed the roles of 'doctor', 'role model', and 'counsellor' during interactions with 'clients' that had the effect of disempowering clients in many cases. The practice of poetry in which peers created couplets in alternating exchanges was identified as one form of naturalistic peer learning that entailed processes of legitimate peripheral participation, meaning negotiation, and reflective learning. Street 'doctory' in which peers provided medical care in the form of procedures, illness discussions, and health consultancy was identified as another naturalistic peer learning pattern involving processes of legitimate peripheral participation, meaning negotiation, and learning through teaching. These findings suggest that naturalistic peer learning involved co-participatory processes that manifested in a diversity of everyday practices. It is recommended that engaging these processes and practices would be useful for interventions, while further research should explore such patterns in other contexts.