Resolving thin layers and accurate delineation of layer boundaries are very important for reservoir characterization. Many seismic inversion methods based on a least-squares optimization approach with Tikhonov-type regularization can intrinsically lead to unfocused transitions between adjacent layers. A basis pursuit inversion algorithm (BPI) based on L1 norm optimization method can, however, resolve sharp boundaries between appropriate layers. Here we formulate a BPI algorithm for amplitude-versus-angle (AVA) inversion and investigate its potential to improve contrasts between layers. Like the BPI for post-stack case (Zhang and Castagna, 2011), we use an L1 norm optimization framework that estimates three reflectivities, namely, Rp, Rs and Rρ. High resolution velocities (Vp, Vs) and density (ρ) can be derived from these parameters by incorporating initial models. Tests on synthetic and field data show that the BPI algorithm can indeed detect and enhance layer boundaries by effectively removing the wavelet interference.