The analysis of S wave polarization from a large number of aftershocks of the 1992 Erzincan earthquake has revealed a clear S wave splitting, which we interpret in terms of seismic anisotropy in and around the Erzincan basin and which concerns at least the first few kilometers of the crust. In the sedimentary basin, our interpretation of the origin of the anisotropy is the alignment of cracks due to the stress field (extensive dilatancy anisotropy (EDA) model) within the deep sediments and possibly below, in the mean direction N335¿. This is rotated by 20¿ from the compressive stress direction N355¿ deduced from other aftershock studies. This difference may either be explained by uncertainties in the stress direction or by a stress distortion related to the activity of some nearby active faults. In the outcropping basement, to the south, the fast S direction is N315¿, significantly different from the present regional stress direction, which implies that the anisotropy is controlled either by a possible, though yet unreported, rock foliation or by cracks related to a local, strongly distorted stress field. Following the preferred hypothesis of stress rotations and of the EDA model, we use a two-dimensional model to calculate the direction of the elastic stress resulting from the combined activity of the adjacent segments on the North Anatolian fault east to the basin (NAE) and within the basin (NAB) with a similar stress release. The resulting trajectories of the maximal compressive stress in the shallow crust are mainly controlled by the NAE segment, owing to its much greater length, and correctly fit the inferred crack directions in the bedrock sites as well as in the basin sites, provided that the stress release on the segments is of the order of the regional deviatoric stress amplitude stored in the shallow crust (1 km in depth). In this model, the stress and crack orientation inferred from shear wave anisotropy can be explained by some slip deficit on the NAB segment with respect to the NAE segment, which is independently suggested by the historical earthquakes in the area. Studies of the crack-induced anisotropy in the shallow crust can thus bring important information on the spatial and possibly temporal variations of crustal strain and stress near active faults, in particular, near the limits of the fault segments marked by jogs, steps, or bends, and might constrain the relative potential of adjacent segments for future earthquake ruptures.¿ 1997 American Geophysical Union