The direction of the slip can change during a rupture episode as a function of time and/or of the position on the fault plane. In this work, we focus on the spatial variations of slip direction during an earthquake. Theoretically, the spatial variations of rake are caused by changes in the direction of the static and dynamic traction. The 1989 Loma Prieta earthquake provides one of the most striking examples of slip direction rotation during the coseismic rupture. All the source models obtained by using different inversion procedures show strong spatial partitioning of slip into strike-slip motion SE of the epicenter and dip-slip motion in the NW section. In this study, we perform both forward waveform modeling and waveform inversion tests to check if the previously observed variations of rake are constrained by seismological data. We use the isochrone approach and a tomographic back-projection algorithm to solve the forward modeling and the inverse problem, respectively. Results show that observed ground velocities can be well reproduced by source models having spatial slip distribution similar to the Beroza's (1991) inversion solution, but having constant rake. Synthetic inversion tests point out that the inversion procedure does not constrain the slip direction. We have analyzed the S-wave polarization in order to constrain the direction of slip along the fault. This analysis confirms that a predominant strike component (rake ∼ 160°) characterizes the slip direction in the southeastern section, while an increase of the reverse component in the northwest section of the fault allows one to explain the observed polarization well. An oblique mechanism is consistent with the fault-plane solution, confirming a spatial variation of the slip direction during the earthquake. If the rake actually varied, then a nonuniform prestress condition (due to a change in fault geometry or to a redistribution of static stress) or a spatially inhomogeneous stress release should be proposed in order to explain the observed changes of slip direction. We conclude that a nonuniform prestress is a likely explanation for the inferred rake rotations. These variations of the direction of slip involve a low value of the initial stress.