Deep-sea ecosystems present difficulties in surveying and continuous monitoring of the biodiversity of deep-sea ecosystems because of the logistical constraints, high cost, and limited opportunities for sampling. Environmental DNA (eDNA) metabarcoding analysis provides a useful method for estimating the biodiversity in aquatic ecosystems but has rarely been applied to the study of deep-sea fish communities. In this study, we utilized pumped deep-sea water for the continuous monitoring of deep-sea fish communities by eDNA metabarcoding. In order to develop an optimum method for continuous monitoring of deep-sea fish biodiversity by eDNA metabarcoding, we determined the appropriate amount of pumped deep-sea water to be filtered and the practical number of filtered sample replicates required for biodiversity monitoring of deep-sea fish communities. Pumped deep-sea water samples were filtered in various volumes (5–53 L) at two sites (Akazawa: pumping depth 800 m, and Yaizu: pumping depth 400 m, Shizuoka, Japan) of deep-sea water pumping facilities. Based on the result of evaluations of filtration time, efficiency of PCR amplification, and number of detected fish reads, the filtration of 20 L of pumped deep-sea water from Akazawa and filtration of 10 L from Yaizu were demonstrated to be suitable filtration volumes for the present study. Fish biodiversity obtained by the eDNA metabarcoding analyses showed a clear difference between the Akazawa and Yaizu samples. We also evaluated the effect of the number of filter replicates on the species richness detected by eDNA metabarcoding from the pumped deep-sea water. At both sites, more than 10 sample replicates were required for the detection of commonly occurring fish species. Our optimized method using pumped deep-sea water and eDNA metabarcoding can be applied to eDNA-based continuous biodiversity monitoring of deep-sea fish to better understand the effects of climate change on deep-sea ecosystems.