Damage detection techniques are the main tool in health monitoring to assess the functionality of structures. Among these techniques, statistical subspace-based damage detection technique is a robust method to evaluate the conditions of a structure without the need of evaluating its modal parameters. This can circumvent all the errors and difficulties in evaluating the modal properties of the structure while the changes in the eigen-structure of the data is identified indirectly. In our previous studies, the effects of damage location, damage ratio, and the noise in the data were investigated. In this study, the sensitivity of this technique to the number of samples of the data is addressed. The interaction of the number of measurements with damage and noise ratio is also investigated. For this purpose a bridge structure located in Reibersdorf, Austria, is considered. This structure is modelled and calibrated to the real test data; subsequently the damage is modelled in one of the elements for different damage ratios and excitation duration. It was demonstrated that this technique can operate robustly even with high noise present in the data by acquiring typical number of measurements especially for reference state data. Moreover, it was concluded that if the length of the reference data would not be large enough, increasing the length of the test data cannot much help in identifying the damage in the structure.