The joint stiffness of jointed part in automobile body structures is one of key factors domination the basic characteristics of body such as strength, a vibration and a collision. The technique to evaluate the joint stiffness using joint stiffness matrix has been suggested. In general, each element of the joint stiffness matrix can't be determined. If the method capable of determining each element of the joint stiffness matrix quickly and accurately is developed, it must be useful in designing body structure. In this paper, a new method to estimate of the jointed stiffness matrix using hierarchical neural networks is described.