Pollution source identification is a secure issue, because it threats the health of hundreds of people. To protect river system from accidental spills, we need accurate and efficient tools to identify such sources. Backward Probability Method (BPM) is one of the tools that gives useful information about the prior position and the release time of pollution. The backward model was originally developed in groundwater; therefore, one of the study's goals is the application of BPM to identify the source location and the release time of the pollutant in surface water. In order to apply the model, a numerical code was developed based on adjoint analysis. It has been tried to use this method in a real case. This model is applied to Severn River. The results show that this model is able to trace the pollution source in a river with natural condition accurately. In the second part, pollution source tracing is conducted in river network based on BPM. Three tests have been designed, and pollution source has been traced very well in all tests. All suspected points are determined with just one backward simulation. The results demonstrate that all suspected points determined by BPM can be a possible source point. This method is so fast and does not need any simplification. Since the simulation is done once, this method effectively reduces computational cost and gives a better choice to decrease the damages. It just needs a limited observation data. Therefore, its cost is the least.