The parametric yield of VLSI strongly depends on the statistical process fluctuations. Thus, the statistical simulation is needed for robust process, device and circuit designs. This article describes the statistical simulation of MOSFETs using Technology CAD (TCAD). Threshold voltage V-th and sheet resistance R-sh Obtained through the statistical simulation were compared with measurements. The mean values of the simulated V-th and R-sh agreed well with those of the experiments within 2% error. The standard deviations sigma revealed sufficient agreements within 30% error, except for n-ch V-th. For a conventional method, larger coverage factors based on the principal component analysis are selected to generate worst-case corner models. We propose a new factor selection method which uses the sensitivity analysis of a small but critical circuit, a ring-oscillator, for obtaining BSIM3v3 worst-case models. In this example, the proposed method selected the factor related to a gate oxidation temperature, whereas the conventional method selected the factor related to a LDD side-wall thickness. The proposed method was applied to the estimations of worst-case performance of the inverter DC characteristics and data-out of DRAM.