Abstract: The sandstone and conglomerate reservoirs in Ulyastai Sag are characterized by strong sensitivity，developed micro-structures，low inter layer stress，low productivity after perforation and the huge difference in effectiveness after fracturing.Precisely and quantitatively selecting the best potential fracturing wells and layers have been the bottleneck in developing the reservoirs.According to the statistic analysis to the data of previous fractured wells，the formation coefficient，porosity and oil saturation etc，as well as displacement amount，proportion of pad fluid and sanding strength，are taken as the main factors influencing fracturing effectiveness.The expert database of the fractured wells is set up.Regression analysis and BP neural network are used to optimally select 3 waiting-for fracturing wells.Practice indicates that there is a nonlinear relationship between the factors and the fracturing effectiveness，the linear regression can not meet the requirement of optimally selecting fracturing wells，the matching error of second order regression and neural network method is 0，predicted error is less than 2%，which can reach the standard of in-situ selecting fracturing wells and layers.

Key words: sandstone and conglomerate reservoir, fracturing, well and layer selection, multiple regression, neural network