Abstract: To improve the construction accuracy of Gaussian radial basis neural network model, a partition real number coding method was proposed by using Gaussian radial basis centers, width, and the connection weights. In this method, the training sample set was used as the network input and output after decoding each individual, and the sum of mean squared error between expectation output of the samples and actual network output was used as evolutionary individual fitness function, and the optimal value of the evolutionary individuals which consist of different hidden nodes was used as the final Gaussian RBF network structure. Simulations were carried out to test the proposed method through two benchmark testing functions. The accuracy of the proposed model construction method was evaluated in terms of different hidden nodes, evolutionary time, number of evolutiongeneration, minimal fitness and the root mean square error. Simulation results showed that the proposed model construction method has high accuracy for different design problems.