Abstract: In order to improve the forecasting accuracy of the neural network model and the computational efficiency,  the structure of Gaussian radial basis neural network based on orthogonal least squares was constructed and the regression models of neural network was given. The center parameters of Gaussian function were determined by the sequence information of the sample point and the connection weights between the hidden layer and output layer was determined by the recursive computation of the orthogonal least squares. The performances of this method and the other literature method used to forecast the model based on chaotic Lorenz time series were compared in terms of forecasting accuracy and the recursive time required. The results indicated that the designed model has many advantages such as higher forecasting accuracy and higher computational efficiency