This study presents a Bayesian damage detection algorithm when structural responses are highly contaminated with noise. A best achievable eigenvector based approach is used under Bayesian framework to detect multiple damages in the structure. It has been illustrated in this study that a false detection of damage may occur when structural responses are highly contaminated with noise. This is particularly observed when less sensitive responses are employed in damage detection algorithm. Therefore, removal of these responses in Bayesian evidence can lead to a better detection of damage. To illustrate this aspect, a 12-story shear building is modeled and the modal data evaluated using eigenvalue analysis and contaminated with high noise levels are used as the Bayesian evidence. The sensitivity of each of these data points is evaluated using the modal data to structural parameters relations. Further, an appropriate data fusion for different modes is established at for damage detection purpose. Results of the study show that the proposed algorithm is efficient enough in damage detection under noisy data.