The purpose of this study is to propose a least-squares method for parameter estimation using Fisher's z-transformation in correlation structure analysis. An advantage of this method in comparison with the unweighted least squares method (ULS) is that the residuals are normally distributed and their variances are homogeneous. A numerical example is given for factor analysis by using the data of Spearman (1904). Through the results of computer simulations in higher-order factor analysis, this method is shown to have less significant errors of parameter estimates than the ULS.