Abstract:Using observation data of underground water level proves to be a comparatively reliable method for inversion analysis of seepage parameters for fractured rocks. Generally,observation data are gained randomly without considering the spatial continuity of underground water level. In this paper, the spatial correlation structure of underground water level is analyzed by using geostatistics. The total estimate errors is used for investigating the result of back analysis for seepage parameters of fractured rock mass under three different sampling scheme .The result shows that optimized sampling scheme can decrease the sampling expenses while insuring the accuracy. The error produced in random samping is bigger than that of optimized sampling scheme when the number of observation data is the same. It is not obvious that increasing the number of random observation data can improve the result of back analysis.