This paper proposes an approach for Web user segmentation and online behavior analysis based on a mixture of factor analyzers (MFA). In our proposed framework, we model users’ shared interests as a set of common latent factors extracted through factor analysis, and we discover user segments based on the posterior component distribution of a finite mixture model. This allows us to measure the relationships between users’ unobserved conceptual interests and their observed navigational behavior in a principled probabilistic manner. Our experimental results show that the MFA-based approach results in finer-grained representation of user behavior and can successfully discover heterogeneous user segments and characterize these segments with respect to their common preferences.