With the development of the micro-electro-mechanical system (MEMS), the flow characteristics in micro-channels have drawn increasing attention. In this paper, numerical simulations are conducted to investigate the flow characteristics of compressible flow through micro-channels and micronozzles. An improved surface roughness viscosity model is used to simulate the effect of surface roughness on micro-channels flow characteristics. Using this model, better agreements between the computational results and the experimental data are found. The result indicates that the surface roughness is one of the important factors affecting the flow characteristics of gas through micro-channels. The numerical investigation on the expansion channel shows that by using the laminar model that considers surface roughness, the computational results and experimental data are consistent when Re<450, whereas deviation increases when Re>450. Based on the synthetic model with considerations of turbulence viscosity and surface roughness, the numerical results and the experimental data are identical. microfluidic devices, surface roughness, compressible characteristics, early transition PACS: 47.61.Fg, 47.60.+i, 47.15.RqWith the development of the fabrication technology of microfluidic devices and systems, there has been great interest in the applications of micro-channels over the years in such diverse engineering areas as silicon-micro-valveless pumps, micro-rocket thrusters, micro-gas turbine generators and bio-analytical instruments. In order to successfully design such devices, the understanding of the flow physics in microscale and their engineering modeling are crucial [1,2]. Wu and Little [3] measured gas flow frictional characteristics for micro-channels with a trapezoidal cross section etched in silicon wafers. The width of the channels was 130 to 200 microns and the depth was 30 to 60 microns. Nitrogen, hydrogen, and argon gases were used in their experiments. They observed that the measured friction factors were higher than the theoretical predictions.