An indirect radial basis neural network (IRBNN) is proposed for improving the accuracy of the approximated functions. The IRBNN is constructed by new prompted functions generated from the Nth order derivative of the approximated function. In this way, high accuracy derivatives in different order can be obtained, so that more accuracy of the numerical results would be given while the IRBNN is employed for creating approximated functions in numerical methods. Numerical results through applications in elasticity show the effectiveness and accuracy of the IRBNN method.