Gas breakdown is one of the key factors limiting the increase of power capacity of the outer surface of high-power microwave dielectric window. It is of great significance to conduct corresponding simulation studies. Compared with the fluid model, the particle-in-cell-Monte Carlo collision model has two advantages. One is that the influence of numerical dispersion and instability problems is insignificant, and the other is that it can accurately describe microphysical processes. Therefore, the breakdown characteristics on the gas side of dielectric window are simulated by using the particle-in-cell-Monte Carlo collision model. The two-in-one macro-particle merging method is introduced into the model, thereby greatly reducing the number of macro-particles tracked. Therefore, the whole breakdown process can be simulated and analyzed. The results show that the spatial and temporal evolution of breakdown under the variable macro-particle weight is in good agreement with that under the constant macro-particle weight. This suggests that the two-in-one macro-particle merging method is applicable under the simulation conditions of interest in this paper, i.e., when the ratio of the effective electric field of microwaves to the pressure is between $1.76\times10^3$ and $1.41\times10^4$ V/(m$\cdot$Torr). Since the yield of the secondary electron emission is much less than 1, gas ionization is the dominant mechanism of breakdown on the gas side of dielectric window. Electron ionization and electron diffusion lead the density and thickness of the plasma to significantly increase over time. The peak of electron density does not appear at the dielectric surface, but at a position of 100–150 μm away from the dielectric surface. This is because a large number of electrons are deposited on the dielectric surface, and the accompanying self-organized normal electric field drives the electrons away from the dielectric surface. Because the pressure of background gas of interest in this work is higher than the critical pressure corresponding to the maximum ionization rate (about 10 Torr), the ionization rate decreases monotonically with pressure increasing, resulting in a slower development of breakdown. The accuracy of the particle-in-cell-Monte Carlo collision model is confirmed by comparing the simulated values of breakdown time with experimental data. This work provides an important theoretical basis for understanding and controlling the breakdown on the gas side of dielectric window. The following figure (a) shows that the mean electron energy under the variable macro-particle weight agrees well with that under the constant macro-particle weight at about 100 Torr. The following figure (b) shows that when the plasma density is increased by a factor of 10⁸, the breakdown process can be considered by using the particle-in-cell-Monte Carlo collision model and a two-in-one macro-particle merging method.