Because high‐resolution transmission electron microscopy (HRTEM) relies on a complex contrast mechanism to produce images of gate dielectric films in cross section, there are many factors affecting the uncertainty of thickness measurements based on these images. A preliminary survey revealed approximately 50 parameters that affect the uncertainty in a gate dielectric dimensional metrology experiment using HRTEM, along with approximately 1,200 two‐term interactions and almost 20,000 three‐term interactions. Using established design‐of‐experiment (DEX) methodologies, I performed a screening experiment based on a2IV(8−4)fractional factorial design to determine which factors had the greatest impact on the absolute error of the thickness measurements. Absolute error was determined by simulating HRTEM micrographs using a multislice calculation. The model used for the simulation consisted of a variable SiO2film approximately 2 nm thick positioned between two pieces of crystalline Si. This approximation to a gate stack was built atom‐by‐atom using commercial molecular modeling software supplemented with custom Tcl scripts to assemble the gate structures from simpler primitives. By varying the molecular model, sample parameters such as crystallographic orientation, film thickness, density, and along‐beam thickness can be varied precisely. Instrument parameters and details of the imaging conditions are inputs to the multislice calculation, a simulation technique that has been vetted by the microscopy community and has been in use for decades. Beam tilt, defocus, and vibration amplitude were the main factors found to have the largest effects, while beam‐tilt↔defocus and defocus↔vibration were the most important two‐term interactions. © 2003 American Institute of Physics