Abstract-The 2004 Indian Ocean tsunami was observed by two satellites, close in space and time, that traversed the Indian Ocean 2 h after the Sumatra-Andaman earthquake, but which observed different tsunami lead wave morphologies. The earlier satellite, Jason-1, recorded a lead wave with two peaks of similar amplitude and wavelength, while the later satellite, TOPEX/ Poseidon, recorded a lead wave with only one longer wavelength uplift. To resolve this disparity, we examine the travel paths of long wavelength waves over the seafloor bathymetry. Waves traveling from the margin will traverse significantly different paths to arrive at the two satellite transects. The result is that the satellites are sensitive to different parts of the margin; Jason-1 is highly sensitive to the margin in the area of the epicenter, while TOPEX is sensitive to a more northerly section. By developing solutions of the ocean gravity wave equations, accounting for dispersion, we show that the double peak of the Jason-1 satellite observations are consistent with coseismic rupture of a splay fault of limited along-strike extent, located north of Simeulue Island. The doubly peaked morphology can be reproduced with co-activation of the subduction zone interface and the splay fault, which creates a seafloor uplift pattern with two distinct areas of uplift. The Jason-1 satellite is sensitive to a splay fault in this portion of the margin, whereas the TOPEX satellite would not be significantly affected by this uplift pattern. By back-projecting satellite observation points to the margin, we constrain the location of the proposed splay fault and find that it correlates with a bathymetric high. The aftershock locations, uplift of corals on Simeulue Island and a fault scarp on Pulau Salaut Besar are also consistent with the activation of a splay fault in the area delimited by the back-projection. Our work also shows that it is critical to fully capture gravity wave dispersion in order to represent features of the lead wave profile that may not be as well characterized by the shallow water (long-wavelength) model. It is also necessary to account for dispersion so as to precisely assess wavefront travel times; this leads us to conclude that the rupture must have reached very near to the trench and propagated with an updip rupture velocity of order 2.0 km/s or more.