In this paper, optimization method for gas cluster generation was studied both theoretically and experimentally for a new dry cleaning process using gas cluster beam. Since the typical size of gas cluster is less than 50 nm, it is expected to cause no damage on the surface. CO2 gas cluster cleaning system was built to evaluate the feasibility of the cleaning by gas cluster beam, which is generated by expansion through the converging-diverging nozzle and chilling. Cleaning test was performed for 25 nm silica particle cleaning on a bare Si wafer, and on a wafer with 80 nm poly-silicon structure. Preliminary result has been obtained, which showed that we could achieve cleaning of contaminants without damage. To optimize the gas cluster cleaning process, we propose the use of particle beam mass spectrometer (PBMS) for the measurement of size distribution and numerical simulation of gas cluster generation process.