Fusion of multiple face and fingerprint matchers based on different biometrics for personal authentication has been investigated in the last years. However, the performance achievable when the expected subject cooperation degree is different from the real one has not yet been sufficiently studied. In this paper, we investigate the performance of several score-level fusion rules when the test set is taken under non-cooperative ("stress") conditions. Results show that fusion allows to increase the robustness of the system under strong changes of the subject?s cooperation degree.