Based on the theory of spline function approximation

a three layer dual-input spline neural network model is constructed and studied

of which the hiddenlayer neurons-activation are the product of spline function with two variables. Based on the space structure of input variable

the grid partition of training data is received. Moreover

network structure can be adjusted with the varying of the training data and the mesh number

and the weight matrix can be determined directly. Simulation experiments show that dualinput spline neural network have high precision

short operation time

and the relationship between the number of training data gridding partition and the network structure is determined.