AIMS: To assess the cost-effectiveness of the CARE-SHAKTI harm reduction intervention for injecting drug users (IDUs) over a 3-year period, the impact on the cost-effectiveness of stopping after 3 years and how the cost-effectiveness might vary with baseline human immunodeficiency virus (HIV) prevalence. DESIGN: Economic cost data were collected from the study site and combined with impact estimates derived from a dynamic mathematical model. SETTING: Dhaka, Bangladesh, where the HIV prevalence has remained low despite high-risk sexual and injecting behaviours, and growing HIV epidemics in neighbouring countries. FINDINGS: The cost per HIV infection prevented over the first 3 years was USD 110.4 (33.1-182.3). The incremental cost-effectiveness of continuing the intervention for a further year, relative to stopping at the end of year 3, is USD 97 if behaviour returns to pre-intervention patterns. When baseline IDU HIV prevalence is increased to 40%, the number of HIV infections averted is halved for the 3-year period and the cost per HIV infection prevented doubles to USD 228. CONCLUSIONS: The analysis confirms that harm reduction activities are cost-effective. Early intervention is more cost-effective than delaying activities, although this should not preclude later intervention. Starting harm reduction activities when IDU HIV prevalence reaches as high as 40% is still cost-effective. Continuing harm reduction activities once a project has matured is vital to sustaining its impact and cost-effectiveness.