Injectable contraceptives are popular in sub-Saharan Africa but have high discontinuation rates due partly to the need for provider-administered re-injection. We compared continuation rates of women who self-injected subcutaneous depot medroxyprogesterone acetate (DMPA-SC) and women who received DMPA-SC from a health-care provider, including community health workers (CHWs).We did an open-label randomised controlled trial based at six Ministry of Health clinics in rural Mangochi District, Malawi. Health-care providers recruited adult women who presented at the six clinics or to CHWs in rural communities in the clinic catchment areas. Participants received DMPA-SC and were randomised (1:1) to receive provider-administered injections or training in how to self-inject DMPA-SC. Randomisation was done via a computer-generated block randomisation schedule with block sizes of four, six, and eight and stratified by study site, generated by an independent statistician. Self-injectors administered the first injection under observation and were sent home with three doses, written instructions, and a calendar. The provider-administered group received a DMPA-SC injection and a calendar, and were asked to return for subsequent injections. Data collectors contacted participants after the 14-week re-injection window at 3, 6, and 9 months to collect continuation data. At 12 months after enrolment or early discontinuation, women had their final interview, which included pregnancy testing. The primary outcome was discontinuation of DMPA-SC, as assessed in the intention-to-treat population. We used Kaplan-Meier methods to estimate the probabilities of continuation and a log-rank test to compare groups. Safety was assessed in the as-treated population, which consisted only of participants who successfully received at least one DMPA-SC injection after randomisation. This trial is registered with ClinicalTrials.gov, number NCT02293694.This study lasted from Sept 17, 2015, to Feb 21, 2017. 731 women underwent randomisation, with 364 assigned to the self-administered group and 367 to the provider-administered group. One woman in the self-injection group withdrew at month 0. Treatment was discontinued by 99 women in the self-administered group and 199 women in the provider-administered group. The 12 month continuation rate was 73% in the self-injection group and 45% in the provider-administered group, giving an incidence rate ratio of 0·40 (95% CI 0·31–0·51; p<0·0001). Adverse events deemed to potentially be treatment-related were reported by ten women (20 events) in the self-administered group and 17 women (28 events) in the provider-administered group. Five serious adverse events were reported during the trial by four women; two events related to DMPA-SC (menorrhagia and anemia requiring hospital admission) were reported by the same woman in the provider-administered group and resolved without sequelae. The other serious adverse events, including one death, were deemed to be unrelated to DMPA-SC. Women who self-injected DMPA-SC had significantly higher rates of continuation than those receiving provider-injected DMPA-SC. Community-based provision of injectable contraception for self-injection in low-resource settings seems to be safe and feasible. Self-administration of DMPA-SC should be made widely available.