Wage subsidies served as a dominant labour market policy response around the world to mitigate job losses in response to the COVID-19 pandemic. However, no causal evidence of their effects exists for developing countries. We use unique panel labour force survey data and exploit a temporary institutional eligibility detail to estimate the causal effects of such a policy—the Temporary Employer/Employee Relief Scheme (TERS)—on job retention among formal private sector employees in South Africa. Using a difference-in-differences design, within the context of an economy with one of the highest unemployment rates in the world, we find that the policy increased the probability of remaining employed in the short term by 15.6 percentage points. Our findings imply that the policy saved 2.7 million jobs during April and May 2020 at a monthly cost of ZAR13,195 (US$1,851 PPP) per job saved. While this cost is large relative to the wage costs of jobs supported by the policy, it compares favourably to more developed country contexts. However, two thirds of the recipients were inframarginal and would have remained employed anyway in the policy’s absence, arguably due to prioritization of rapid disbursement of relief over accurate targeting. We additionally examine heterogenous effects by subsidy intensity and find that effects are positive but marginally regressively distributed across the subsidy distribution.