Using design of experiment and data transformation, the model of thermal oxidation furnace is developed. The model involves six input factors and two responses. Only 16 run experiments are needed through the fractional factorial experimental design. The Box-Cox transformation method is used to transform the experimental data, and the optimum transformation style is determined. The responses are transformed so that the residuals meet the assumption of ANOVA and the model can explore more information here. After responses transformation the figure of merit for fitting, Adjust-R 2 , is improved from 93.54% to 98.64%. The models are subsequently used to optimize the thermal oxidation process. The non-uniformity of film is improved from 0.2% to 0.08% with the other specification satisfied.