Although biometric recognition systems provide many advantages over traditional recognition methods, they can be vulnerable to specific attacks which may considerably decrease their security. In this paper we focus on the hill-climbing attack which is peculiar of biometric systems. Specifically, we evaluate the effectiveness of general approaches relying on parametric functions optimization for performing this kind of attack, and propose possible countermeasures which can be used for increasing the system robustness without significantly affecting its recognition performance. An application to on-line signature biometrics is taken into account to test both the proposed attacks and some possible countermeasures.