Abstract: Selecting rational hidden structure is very important question on multi-layer neural network in application. Applying optimum algorithm of neural network and algebra equation theory, nonlinear equations, expressing the relationships between computing output and ideal output of neural network were developed. The immanent connection of weights and threshold variables, standard sample numbers and node numbers of output layer are suggested. Hidden structure basis equation is given. A kind of analyzing method of rational construct of multi-layer neural network is proposed. Direct and indirect computing method were studied to compute quantitatively the numbers of hidden layers and the unit numbers per hidden layer. The analysis validity was showed through comparison of rational structure with various passible structures of neural network examples.