Dissolved oxygen (DO) is a very important indicator for river ecosystem health. Taking the DO test in the Xiaogan Reach of the Lunhe River as a case, DO prediction model based on grey LM-BP network of random Genetic Algorithm (GA) was put forward and established. The raw data of DO were predicted based on grey theory, and the residual data were fitted and tested in random GA-LM-BP network. As it turned out, the data rate of accordance is approximately 100% based on random GA-LM-BP network. Furthermore, the last three months data were tested and the errors were less than 2.933% . Thus, the grey LM -BP network of random GA could be used to predict DO in local area. Therefore, it can provide a new methodology for the analysis of water quality.