Abstract Using a 2-D plane stress finite-element model with elastic and elasto-plastic rheologies, appropriate for deformation within the brittle upper crust, we analyse the relationship between kinematics of convergence, deformation and stress distribution in the present Taiwan collision occurring between the Ryukyu and Luzon subduction zones. The distribution of stress trends calculated in our models is compared with a synthetic map of actual stress trajectories based on the most recent data available in the collision zone. These data combine present-day sources (from borehole breakouts and earthquake focal mechanism) with the reconstruction of Quaternary palaeostress (from fault slip data analyses), resulting in a complete map of compressional stress trajectories which is used to constrain our models. We show that the distribution of stress trajectories in the active Taiwan collision is principally controlled by: (1) the geometric configuration of the boundary between Eurasia and the Philippine Sea plate; (2) the shape and rheological properties of major structural units; (3) the direction of convergence of the Philippine Sea plate relative to Eurasia; and (4) the influence of the opening of the Okinawa Trough. The study of a two-dimensional elastic and elasto-plastic finite-element modelling of the subduction-collision in and around Taiwan allows us to estimate the influences of these different parameters in the stress pattern. Taking into account both the simplifying assumptions of the numerical modelling and the angular uncertainties of field determinations, the fit between the calculated stress pattern of the finite-element model and that determined based on the geometrical synthesis of field analyses is rather good in general, indicating that our model is valid to first approximation. Misfits remain minor and can be explained by data uncertainties and simplifying modelling assumptions (for instance, the shape of the corner of the collision zone is critical but is not accurately known; also limited decoupling in the Longitudinal Valley collision zone was not considered in our models although it certainly plays a role). Some interesting features of our model are: (1) the greater influence of the shape of the collision zone in comparison with that of the direction of convergence; (2) the requirement for a trench retreat related to suction force in the Ryukyu Arc; and (3) the crucial role of the interaction between Okinawa Trough opening and collision at the sharp northwestern corner of the Philippine Sea plate including its influence on the geological evolution of northeastern Taiwan.