This study is based on the geometrical distribution of hypocentres offshore northeastern Taiwan, where the Luzon volcanic arc collides against the Eurasian margin. The analysis of deep earthquakes (50-300 km) provide geometrical constraints on the shape of the Philippine Sea (PS) seismogenic slab. At 123 degrees E, the slab dips 60 degrees + or -10 degrees to the north, whereas near its western edge, between 121.5 degrees E and 122 degrees E, it dips 70 degrees + or -10 degrees to the northeast. The slab is deformed, torn in a concave shape to the north and upward, and thus becomes shallower to the west. Furthermore, the sharp termination of the seismic activity to the west probably marks the western boundary of the PS slab. The study of shallow earthquakes (0-50 km) reveals two major earthquake swarms. One of these swarms is located in the northern colliding Luzon arc, with earthquakes occurring down to 50 km depth. Focal mechanisms are compatible with either a 25 degrees -dipping, SE verging thrust, or a 65 degrees -dipping, NW verging thrust. The velocity of convergence between the northern part of the Coastal Range (the northernmost segment of the Luzon arc) and the stable Chinese platform is only 1 to 2 cm/yr north of 23 degrees 40'N, as compared with a convergence rate of about 6 cm/yr for the southern part of the Coastal Range. A NW-dipping thrust develops within the PS plate just east of the northern tip of Luzon arc as observed on field and marine investigations. The second swarm of earthquakes is observed in the northern edge of the Nanao forearc basin where epicentres occur within a 60 km long, NW-SE trending zone. Vertical sections crossing this seismic swarm show that seismicity is distributed between 0-50 km, and not restricted to plate interface. Focal mechanisms in this earthquake swarm present consistent nodal planes representing either a gently northward-dipping (15 degrees ) thrust (with P-axis trending N-S, i.e., deviated about 35 degrees from the convergence vector) or a high-angle (80 degrees ) southward-dipping thrust. The first solution, consistent with a typical interface seismogenic zone, fails to explain the geometrical distribution of earthquakes. Because quakes distribute along a subvertical plane and no transcurrent faulting is observed along the Ryukyu arc, we favour the second solution. Such high-angle reverse fault within the subducting plane could represent the tear fault previously evidenced from other marine data.