This research presents a new agent-based model, developed within the GAMA platform, to simulate the impacts of two tsunami warning channels (sirens and/or cell-broadcast) on massive pedestrian evacuation. The study area is located in Cannes, in the south of France, where previous empirical studies provided information on milling time, hazard perception and audibility of sirens. A series of simulation experiments were conducted to measure the effects of inputs (e.g. the number of agents or human density) on the number of agents who joined safe places, and to estimate the impact of various strategies (siren or cell-broadcast activation, in time or with a delay of 15 min), as well as combinations of these. The results confirmed that a quick alerting strategy increased the number of agents who were safe in time, while delays in warning activation decreased this simulation output data. However, weak hazard perception renders the evacuation process less efficient, even if Cell-broadcast and/or sirens are quickly activated. These early findings could then help stakeholders to determine the most suitable strategies to be applied in short term and at local scales.