Background: Understanding population genetic structures is crucial for planning and implementing conservation programmes to preserve species’ adaptive and evolutionary potential and thus ensure their long-term persistence. The grey mullet (Mugil cephalus) is a globally distributed coastal fish. Its populations in waters surrounding Taiwan on the western Pacific fringe are divided into at least two stocks (migratory and residential), but questions remain regarding their genetic divergence and gene flow. Methods and Results: To cast more light on this, allozyme variations at 21 presumptive gene loci of 1217 adult grey mullets from 15 localities in Japan, Taiwan and mainland China, and four gene loci from 1470 juveniles from three localities in Taiwan were used to investigate patterns of genetic variation. The mean expected heterozygosity (He) was 0.128—ranging from 0.031 (Matsu) to 0.442 (Kaoping)—and the mean observed heterozygosity (Ho) was 0.086—ranging from 0.017 (Kaohsiung) to 0.215 (Kaoping). Both AMOVA and the high overall mean FST of 0.252 indicated enormous genetic differentiation among populations and the positive mean value of FIS was 0.328, indicating a deficiency of heterozygotes. PCoA indicated that the samples of M. cephalus could be split into three groups and STRUCTURE analysis showed that all individuals were grouped into three genetic clusters. The results of mutation-drift equilibrium tests did not suggest that the populations experienced any recent genetic bottleneck. The results from all localities in the present investigation showed significant change in the GPI-A genotype frequencies with latitudes—e.g., increases in GPI-A*135/135 homozygote frequencies and GPI-A*100/100 frequencies were highly correlated with latitudinal cline. All migratory populations with the GPI-A genotype were almost exclusively the GPI-A*100/100 homozygote. During the life history of M. cephalus, the GPI-A*100/135 heterozygote frequency significantly decreases with age. Conclusions: Based on these data, we suggest that each GPI-A genotype represents trait combinations of higher fitness in some portions of the environment. Furthermore, the genotypic frequencies change in accordance with life stages, suggesting that selection occurs throughout the life span.