Based on the reference to domestic and foreign correlative theories and methods, cost-significant theory and neural network theory are used to estimate project cost in the paper. The cost-significant theory is put forward to solve the tedious operation issues by finding out significant items to simplify the operational difficulty of engineering cost estimation. Then the BP neural network is applied to distill the data of CSIs and csf from the completed projects. It has realized the accurate prediction of project investment by using the two nonlinear theories. The basic theories of CS and BP neural network are illustrated by an example From the example, it shows that the relative errors are so small that they can meet the accurate demands of cost estimation. Meanwhile, the test results show that the model based on cost- significant theory and neural network theory is successful and effective for practical engineering.