This article presents an experimental study of privacy-related attitudes and behaviors regarding a music recommender service based on two types of user modeling: personality traits and musical preferences. Contrary to prior expectations and attitudes reported by participants, personality traits are frequently disclosed to the system and other users, indicating that embedded modeling of user personality does not represent an acceptance barrier. Discrepancies between privacy attitudes and behaviors have been reported before in the context of e-commerce applications, but the corresponding studies could not exclude several conflicting hypotheses, such as participants expressing attitudes outside the context of specific privacy dilemmas and contact with researchers, which may have mitigated perceived privacy risks. Arguably, these are fundamental problems in empirical investigations into privacy that apply to most published works relating to privacy and user modeling. Measures to control these factors in this study are discussed, and methodological suggestions for future research are presented.