The work aims to study the building and evaluation methods of calibration regression equation for methylbenzene in solvent residues in order to improve the test level of solvent residues. Based on the analysis method for solvent residues in GB/T 10004—2008, the weighted least square method of methylbenzene was adopted to establish the calibration regression equation. The weighted calibration regression equation was evaluated systematically and comprehensively through coefficient of determination, standard deviation, statistical test, confidence interval, control limits, linear range and uncertainty. The practical regression effects of calibration regression equations obtained in ordinary and weighted least square methods were compared. The weighted calibration regression equation model was appropriate and of significant linearity and remarkable fitting degree. It could effectively eliminate the effects of heteroscedasticity on the calibration regression equation, reduce the relative error in low-concentration determination significantly, and ensure the accuracy and reliability of the test results. The proposed method can be important guide for the building and evaluation of calibration regression equation, quality control and processing and analysis of data in the laboratory.