I have analyzed teleseismic body waves from the 2003 Tokachi-oki earthquake (M w 8.1), and inferred the slip distribution. Two simple fault models are assumed for estimating the effect of fault geometry on derived slip distributions. One is a single planar fault with a dip of 20° and the other is a compound fault having a shallow plane with a dip of 5° and deeper, landward plane with a dip of 20°. The compound-fault model is preferable because it explains the initial part of the observed P-waves better. It is found that the planar fault has one asperity (patch of large slip) near the hypocenter and the other asperity to the landward side of the hypocenter. The compound-fault model shares the landward asperity with the planar-fault model, but does not have the asperity near the hypocenter. The other asperity on the compound fault is found far from the hypocenter. This difference of the slip distributions suggests the importance of accurate modeling of the fault dip angle when deducing the slip distribution from teleseismic body waves.