We present an information-theoretic discussion of authentication via graphical passwords, and devise a model for entropy estimation. Our results make face-recognition based authentication comparable to standard password authentication in terms of uncertainty (Shannon-entropy) that an adversary is confronted with in both situations. It is widely known that cognitive abilities strongly determine the choice of alphanumeric passwords as well as graphical passwords, and we discuss various selected psychological aspects that influence the selection process. As a central result, we obtain a theoretical limit to the entropy of a face-recognition based authentication in the light of some social engineering techniques (dictionary attacks on graphical passwords). Remarkably, our results hold independently of any information that can be obtained from the internet or through other forms of social engineering. Thus, we obtain very general bounds on the quality of authentication through face-recognition that solely depend on the authentication mechanism.