The reliability demonstration analysis of intaking water from riverside water resource requires the analysis of river runoff.In order to enhance the accuracy of predicting runoff of the river,the paper proposed a method of predicting river runoff state by using grey-markov model based on initial value optimization.the method took series data of the first duration, data of the last duration and data before the last duration as the initial condition.The prediction results were weighted average.The optimized GM (1,1) model was established and the grey prediction results were modified by the markov model.The model was utilized to compare the runoff predicted results and runoff actual results of Juliu river hydrological station of Lliaohe river from 2000 to 2010 in January and February, the prediction precision was high. It proved that the grey-markov model of initial value optimizing is feasible and the prediction is more accurate.The paper used the model to predict the dry river runoff in future 10 years and analyzed the reliability of water intake from a riverside water resource which will be constructed and provided the technical basis for the water resources argumentation of proprosed water intaking engineering.