Delegating the computation of a polynomial to a server in a verifiable way is challenging. An even more challenging problem is ensuring that this polynomial remains hidden to clients who are able to query such a server. In this paper, we formally define the notion of Private Polynomial Evaluation (PPE). Our main contribution is to design a rigorous security model along with relations between the different security properties. We define polynomial protection (PP), proof unforgeability (UNF), and indistinguishability against chosen function attack (IND-CFA), which formalizes the resistance of a PPE against attackers trying to guess which polynomial is used among two polynomials of their choice. As a second contribution, we give a cryptanalysis of two PPE schemes of the literature. Finally, we design a PPE scheme called PIPE and we prove that it is PP-, UNF-and IND-CFA-secure under the decisional Diffie-Hellman assumption in the random oracle model.