Abstract：Initial geostress is one of the most important factors in the design and construction of the geotechnical engineering. According to the measured geostress data，there are many methods of calculating initial geostress field with different advantages. According to the measured geostress data from Xiluodu Hydropower Project，through 3D FEM combined with the linear multivariate regression analysis principle，the neural network as well as the genetic algorithms，the geostress field back analysis are conducted. It is shown that the back analysis results by the three kinds of methods are very close，and all the methods can simulate the real geostress field. The further analysis shows that，since the yield zone in the rock mass is small，and nonlinear character of the initial geostress field of dam site is faint；the linear multivariate regression analysis results are acceptable. The results also show that，under the condition of accurate results，the linear multivariate regression analysis is of more merits，that is，more convenient，rapider，easier，only solution，compared with the neural network method and the genetic algorithms.
