Abstract: The methods for bridge damage identification are usually of low precision due to measurement errors, random environmental factors and model errors. Fundamental theories of statistics and random analysis are used to develop a new method that identifies damages such as cracking of concrete bridges. The proposed method firstly identifies the main factors that affect the monitoring response magnitude and regards them as the major cause. A statistical model is then established to relate the magnitudes of response and cause used to predict the health of a structure. Using this method uncertainties such as random environment factors, measurement errors and so on can be effectively considered, and the calculative model error in structure analysis can also be avoided. The precision of damage identification can be effectively improved. Model tests of two concrete beams demonstrate the effectiveness of the proposed model.