A method of applying the statistical regression analysis to thermal network correction is investigated. The reciprocal relations imposed on the parameters in the network model are taken into account by use of the Lagrange multipliers. Then a simple form of the normal equation of the parameters is obtained. An equation for the multipliers is derived and shown to be easily solved. So that the method is applicable to actual and large scale problems. The stepwise procedure of Efroymson is used for the regression analysis of the temperature data. Numerical study on a sample network showed effectiveness of the F-test in the procedure for examining the statistical importance of each parameter. It is found that there exists an optimum value for the critical F. When the only parameters contained in the original model are given as unknowns, the optimum value is about one, however, when all the possible parameters are given as unknowns, the value is about two. This result agrees with the conclusion of a general theory of optimum model estimation.