Influenced by numerous physical factors, the evolution of fog droplet size distributions (DSDs) during the fog lifecycle is not yet fully understood and difficult to represent realistically in numerical models, constraining the accuracy of fog forecasting. To improve understanding of the fog evolution, field observations under a polluted background were conducted during the winters of 2006&ndash;2009 and 2017&ndash;2018 in Nanjing, China. Among the 27 observed fog events, microphysical properties such as fog number concentration (<em>N_f</em>), liquid water content (LWC), volume-mean radius (<em>R_v</em>) and effective radius (<em>R_eff</em>) vary substantially. The unimodal (3 &mu;m), bimodal (3, 21 &mu;m) and trimodal (around 3, 13, 21 &mu;m) DSD were observed. As the fog developed, the DSDs evolved from unimodal to multimodal. The third mode centered at 13 &mu;m in trimodal cases appeared after the other two modes, typically around the time LWC reached its maximum, corresponding to the mature stage of fog. For all mode types, the probability density function decreased with increasing <em>N_f</em> and LWC. <em>R_v</em> is generally greater than 4  &mu;m and <em>R_eff</em> greater than 6 &mu;m for trimodal DSDs. Based on the observational findings, a segmented gamma fitting was applied to the mean DSD with partition points at 10 and 21 &mu;m. Comparison between microphysical parameters derived from the fitted DSD and those from observations indicates that the three-segment fitting provides more accurate estimates of <em>N_f</em> and LWC. Moreover, the three-segment gamma fitting substantially improves the representation of <em>R_eff</em>, absorption coefficient and optical thickness, with most deviations constrained within 20 %.