A 3D-VAR assimilation system was established to assimilate the observations of PM_2.5 and PM₁₀, and assimilation and forecast experiments were performed during Nanjing Youth Olympic Games (NYOG). The control variables of this assimilation system were PM_2.5 and PM_2.5~10 (that was the rest of PM₁₀ after taking out PM_2.5). The background error covariances of PM_2.5 and PM_2.5~10 were estimated by using the simulated products of WRF-Chem of August 2014 in Nanjing. The results showed that the decreases of correlation coefficients of PM_2.5 with the distance in the horizontal and vertical directions were less than those of PM_2.5~10, with the possible reason being that the particle size of PM_2.5 was smaller, the life cycle of it was longer and it spread further in the atmosphere. In addition, the WRF-Chem model was run with assimilation during NYOG (from August 16^th to August 28^th, 2014), by using the hourly data of PM_2.5 and PM₁₀ observed from 134 measurement sites around Nanjing. Evaluated with the observations in the innermost of the model area, the experiment results suggested that the aerosol forecasts of the initial fields can be significantly improved by the assimilation. The correlation coefficients of PM_2.5 and PM₁₀ increased by over 53%, the root-mean-square errors of the two reduced by over 55%, and the biases reduced by about 90%. The following aerosol forecasts in positive effect can be obviously improved by the assimilation and the benefit from the assimilation of aerosol can last more than 20hours. The forecast of PM₁₀ was better than that of PM_2.5 by the model.