Parameters of microscopic traffic simulation model could be calibrated using the mean error as goodness of fit. However, due to the unmatched specification of model with the granularity of data and lack of regard for the individual information, optimization algorithm might be misled to find unreasonable solutions. To deal with such issue, mesoscopic model was chosen for its simpler specification and the goodness of fit function based on the distribution of singlevehicle speed is proposed in this paper. Then, a case study was employed for calibration of simulation model by using the network of inner ring road, Guangzhou. Moreover, a traditional calibration method was used for comparison and analysis. Results show that the new measure of goodness of fit outperform the traditional one in terms of the calibration effect, parameters rationality and optimization efficiency, which reveal that the proposed method has the potential to be feasible and popularized.