In this paper, a practice of design of experiments and data analysis for 28nm technology NMOS devices was introduced. Optimal design with I-optimality criterion was used for 5 input variables with 3 levels for each device, where total 23 runs/wafers were generated. Based on the response surface models by an excellent regression on device parametric data, both automatic maximum desirability analyses by JMP software and engineering knowledgebased analysis were carried out. The two methods generated similar prediction and the latter one helps gaining more insight on the device optimization. The prediction was exactly reproduced in silicon experiment.