As a result of aging under environmental conditions, highway bridges are gradually deteriorating. This issue has gradually become a worldwide concern. Vibration-based structural damage detection has attracted more and more attention over the past decade. However, the effects of changes in environmental conditions have been recognized as main barriers against the application of vibration based damage identification techniques for real-world bridges. In this thesis a statistical method is presented for highway bridge damage detection by applying hypothesis-testing theory in conjunction with dynamic linear regression model. This method makes improvement by considering the environmental effects in damage detection. Thus changes of indicator caused by damage can be separated from those caused by environmental conditions. As a result, the use of this method can avoid misclassification for damage detection results. In order to account for effects of changes in environmental conditions on structural damage detection results, a statistical damage detection procedure is presented. It is assumed that the dynamic responses from both healthy (undamaged) state and damaged state are available. The presented method consists of (1) applying a dynamic linear regression model, auto-regressive with external inputs, to model the relationship between damage signatures and time varying environmental conditions, and (2) performing damage detection based on hypothesis-testing. The effectiveness and robustness of the presented procedure is demonstrated by damage detection of a simply supported beam with different level of damage severity. The research work presented in this thesis contributes to apply statistical techniques for the development of robust and reliable vibration-based structural health monitoring systems for real-world structures.