New magnetic anomaly identifications W of the Shackleton Fracture Zone show 5 spreading sections, separated by fracture zones. In the 2 most southerly, the ridge crest collided with a trench at the margin of the Antarctic Peninsula only 6.5 and 4 Ma ago, the latest of a series of collisions starting at the base of the peninsula SO Ma ago. Following each collision, spreading and subduction both stopped. Opposite the South Shetland trench and actively extending Bransfield Strait, spreading also stopped 4 Ma ago, but before the spreading centre reached the trench. The tendency of the subducting plate to continue sinking probably initiated the opening of the Bransfield Strait. The parallelism between fracture zone orientation on the descending plate and convergence direction, previously thought responsible for the tectonic segmentation of the peninsula, was effective for only 10–30 Ma before collision. Before that, either the fracture zones did not extend to the trench, or the subducting ocean floor formed at the Farallon–Phoenix (Nazca–Aluk) boundary, with fracture zones oblique to the subduction direction. The Aleutian Arc–Kula Ridge model, in which arc magmatism virtually ceases when ocean floor younger than 25–30 Ma occurs beneath the arc, fits the distribution of Antarctic Peninsula radiometric dates, explaining a 50–60 Ma gap between collisions and youngest ages in the S, and possibly the migration of the youngest activity towards the trench. Thus, gaps in the geologic record of arc magmatism need not imply cessation of subduction. The progressive steepening of the peninsular margin towards the 4 Ma collision site suggests tectonic erosion of the fore-arc as the ridge crest approached.