Abstract: While drilling with PDC Bit,Some problems are come out when using the methods of traditional recorder well.Lithology identification is crucial.The data collected by mud-logging equipment,for examples,ROP,WOB,torque et al.,reflects the different sides of rock’s property in subsurface.The relation among them is complex and BP neural network is adopted in three layers.The section of Zhang 2104 well in Zhangdian oil field is used as predictive model and the depth is between 2500 and 2700 meters.In order to test the precision of the model,the section of Zhang 2104 well(2000-2200 m) and zhang 2201 well(2600-2800 m) are predicted.The precision predicted to Zhang 2104 is 93.6% and zhang 2201 well is(89.2)%,as the pure rock is almost 100%.The error remains the identification to the rock filled with nearby 50% clay.The result presents that this method is practical.