We present the results of a detailed shear wave splitting analysis of data collected by three temporary broadband deployments located in central western South America: the Broadband Andean Joint experiment (BANJO), a 1000-km-long east-west line at 20 ¿S, and the Projecto de Investigacion Sismologica de la Cordillera Occidental (PISCO) and Seismic Exploration of the Deep Altiplano (SEDA), deployed several hundred kilometers north and south of this line. We determined the splitting parameters ϕ (fast polarization direction) and Δt (splitting delay time) for waves that sample the above- and below-slab regions: teleseismic *KS and S, ScS waves from local deep-focus events, as well as S waves from intermediate-focus events that sample only the above-slab region. All but one of the *KS stacks for the BANJO stations show E-W fast directions with Δt varying between 0.4 and 1.5 s, However, for *KS recorded at most of the SEDA and PISCO stations, and for local deep-focus S events north and south of BANJO, there is a rotation of ϕ to a more nearly trench parallel direction. The splitting parameters for above-slab paths, determined from events around 200 km deep to western stations, yield small delay times (≤0.3 s) and N-S fast polarization directions. Assuming the anisotropy is limited to the top 400 km of the mantle (olivine stability field), these data suggest the following spatial distribution of anisotropy. For the above-slab component, as one goes from east (where *KS reflects the above-slab component) to west, ϕ changes from E-W to N-S, and delay times are substantially reduced. This change may mark the transition from the Brazilian craton to actively deforming (E-W shortening) Andean mantle. We see no evidence for the strain field expected for either corner flow or shear in the mantle wedge associated with relative plate motion. The small delay times for above-slab paths in the west require the existence of significant, spatially varying below-slab anisotropy to explain the *KS results. The implied anisotropic pattern below the slab is not easily explained by a simple model of slab-entrained shear flow beneath the plate. Instead, flow induced by the retrograde motion of the slab, in combination with local structural variations, may provide a better explanation. ¿ 2000 American Geophysical Union