PKP(Cdiff), the PKP diffraction waves after PKP-C cut-off, was identified in the PKIKP (or PKP(DF)) wave coda of teleseismic seismograms from a deep earthquake in western Brazil. An array stacking technique was applied to the short-period waveforms from a regional seismic network in Taiwan to retrieve the weak PKP(Cdiff) waves between 159¿ and 162¿. The differential travel times and slowness of PKP(Cdiff) and PKP(DF) were used to constrain the P-velocity structure near the inner-outer core boundary (ICB). The results of this study demonstrate that in some cases the ICB structure can be constrained by data from a single event using presently available short-period regional seismic data and applying suitable array techniques. With a correction for axi-symmetric anisotropy applied to the inner core. the results of this study agree with the PKP properties of model VMOI, a regional ICB model revised from PREM covering the ICB sampled area of this study, and model AK135, a new proposed earth model. The C-cusp position was reported at 151.4¿ from this event. Theoretical predictions for another new proposed earth model SP6 is in significant disagreement with the observed differential travel times between PKP(Cdiff) and PKP(DF). At a distance of 159¿, the residuals (observation-prediction) included the observation errors (¿0.2 sec) are approximately +0.17 sec, -0.31 sec and +0.69 sec for VMOL, AK135 and SP6, respectively. Neither model predicts a significant differential slowness residual. This misfit along this path may be the result of the lateral heterogeneity of the sampled region near the ICB relative to both spherically symmetric models or the complications imposed by the non-axi-symmetric anisotropy of the inner core. ¿ American Geophysical Union 1996