For Multilook PolSAR data classification, measuring the (dis)similarity of two covariance matrices is an important task. The classical Wishart distance and its variants have been successfully used in many applications. These distribution based distances perform better than the lp norms defined in Euclidean space. Since the covariance matrices form Riemannian manifold, the geometric metrics defined on it are more suitable for similarity measure. We introduced the Jensen-Bregman LogDet Divergence (JBLD) into PolSAR data classification. JBLD has several desirable properties and is easy to calculate. Experiments on DLR (German Aerospace Center) F-SAR and CETC-38 Institute (China Electronics Technology Group Corporation) airborne PolSAR data show that the unsupervised classification can benefit much from this well-defined distance.