Abstract: In order to detect the bridge damage caused by train load, a method for damage location identification was put forward based on acceleration response caused by vehicles. With this method, structural damage vulnerability is firstly analyzed to find out vulnerable sections. From the damage states of these sections, several subdomains are selected from the time domain when a train is running. In every subdomain, the damage states of the studied sections are supposed to remain the same, and damage location identification of every subdomain is divided into 2 hierarchies, a model for damage location identification is established for each hierarchy by using the acceleration history data as damage indexes, sample sets are optimized from several viewpoints, and support vector machine is taken as a tool of classification. The analysis results for a continuous girder show that by taking the correlation between structural damage states and train load into account, this method can detect the vulnerable sections in the time subdomains when damage is prone to appear, and obtain the preferable results of damage location identification with only a little change under low-level noise disturbance.