Rocky reef habitats are critical for supporting exceptionally high fish species diversity, a vital element of biodiversity. However, monitoring fish species in these habitats using traditional fisheries resource survey methods poses significant challenges. Therefore, there is an urgent need to develop sensitive and efficient methods for biodiversity monitoring. Four sampling stations were strategically selected around the Miaozihu Islands within the Zhongjieshan archipelago, and surface seawater samples were collected across four seasons (February, May, August and November in 2019). The spatial and temporal distribution of fish was analyzed using environmental DNA metabarcoding technology. The findings revealed a total of 37 fish species across 10 orders 26 families 36 genera. The seasonal abundance of fish species followed a trend of summer >winter > autumn > spring. Remarkably, only 2 fish species were observed in all four seasons, representing approximately 5.41% of the total species count, while 54.05% of the species were detected in a single season only. The non-metric multidimensional scaling (NMDS) analysis and the analysis of similarities (ANOSIM) test indicated significant variation in fish composition among seasons (P<0.05), but not among stations. The fish community's uniformity remained relatively stable. Although the diversity and richness indices did not exhibit significant differences among stations, they peaked in summer and troughed in spring. Overall, the distribution of fish in the coastal waters of the Zhongjieshan Islands is significantly influenced by the seasonal changes rether than station locations. This study introduces an innovative, environmentally friendly approach to fish diversity assessment in marine environments akin to those of the Zhoushan islands and reefs, offering valuable technical insights for the management and conservation of fisheries resources.