The aim of this work is to infer the slip distribution and rupture velocity along the rupture zone of the 26 December 2004 Sumatra–Andaman earthquake from available tide gage records of the tsunami. We selected waveforms from 14 stations, distributed along the coast of the Indian Ocean. Then we subdivided the fault plane into 16 subfaults (both along strike and downdip) following the geometry and mechanism proposed by Banerjee et al. (2005) and computed the corresponding Green’s functions by numerical solution of the shallow-water equations through a finite- difference method. The slip distribution and rupture velocity were determined simultaneously by means of a simulated annealing technique. We compared the recorded and synthetic waveforms in the time domain, using a cost function that is a trade-off between the L1 and L2 norms. Preliminary tests on a synthetic dataset, together with a posteriori statistical analysis of the model ensemble enabled us to assess the effectiveness of the method and to quantify the model uncertainty. The main finding is that the best source model features a nonuniform distribution of coseismic slip, with high slip values concentrated into three main patches: the first is located in the southern part of the fault, off the coast of the Aceh Province; the second between 6.5° N and 11° N; and the third at depth, between 11° N and 14° N. Furthermore, we estimated that the rupture propagated at an average speed of 2.0 km/sec.