Swaziland has the highest prevalence of human immunodeficiency virus (HIV) in the world. Attrition (loss to follow-up and mortality) in people living with HIV/AIDS (PLWHA) already on treatment is a major challenge, undermining achievements of the antiretroviral treatment (ART) programme in Swaziland. The contributing factors to attrition in the Swazi context are unclear. This study aims to (1) estimate attrition from the ART programme 12 months after ART initiation in Swaziland, and (2) determine the predictors of attrition in PLWHA treated with ART in Swaziland. A retrospective cohort study using national baseline data was conducted. A competing-risk Cox proportional hazard regression was used to determine the predictors of attrition. We estimated 10·3% (95% confidence interval 10·1–10·6) attrition in 16 423 participants that initiated ART in 2012. Attrition was significantly associated with sex, age, district, treatment supporter at initiation, co-infection of HIV and TB, functional status, WHO clinical stage, and ownership of facility. Our study can form a base of policies, plans, and service delivery strategies for preventing and controlling attrition in Swaziland.