This study presents two new strategies for real-time control of a metro (rail transit) system. Both act upon the holding times of trains at stations and attempt to minimise passenger wait times. The first strategy applies heuristic rules and requires very few computational or infrastructure resources. The second strategy is based on predictive models (MPC) and numerical optimisation of an objective function using genetic algorithms, and requires online measurement of state variables. The two strategies are compared to an open-loop control base case that imposes constant holding times. Testing is conducted by a dynamic simulator calibrated with real-world data from the Valparaiso (Chile) metro system. The simulations employ origin–destination matrices and assume finite train capacity and minimum security headways between trains. The results indicate that the simple strategy produces improvements of 32.7% in wait times and 35.5% in travel times compared to the open-loop case. The model predictive control (MPC) strategy reduces wait times by 24.0% and travel times by 5.5% compared to the simple strategy. Given the high costs of MPC infrastructure, the authors conclude that for the situation studied, an economic cost–benefit analysis must be performed before choosing one or the other approach during a real implementation.