As environmental DNA (eDNA) metabarcoding has become widely used to monitor species-level biodiversity, research has begun to apply this approach to monitoring genetic diversity within species. However, the practical application of eDNA metabarcoding to intraspecific diversity monitoring remains challenging due to spurious amplicon sequence variants (ASVs) present in the data, even after careful data processing with existing filtering and denoising (error correction) methods. To address this issue, we developed a novel, statistical model (Gaussian mixture model)-based ASV filtering method, gmmDenoise. As we have reported elsewhere that gmmDenoise showed high performance in benchmarking tests using single-species (ayu fish) eDNA metabarcoding datasets, this study presents how gmmDenoise can be used to derive reliable intraspecific diversity estimates and population genetic inferences from more common multi-species eDNA metabarcoding data. Specifically, using two publicly available fish eDNA metabarcoding datasets (one obtained from riverine waters and the other from estuarine waters), we will demonstrate that gmmDenoise can effectively filter unreliable low-abundance ASVs in amplicon data. Furthermore, we show that the gmmDenoise-filtered ASV data can provide significant information about species’ population genetic structure, revealing in two different species (the stream-dwelling Japanese torrent catfish and coastal dwelling flathead grey mullet) similar phylogenetic and phylogeographic patterns to those found in biological sample-based studies. Based on these results, we argue that our gmmDenoise can be an alternative or supplementary option to existing ASV filtering and denoising methods. The gmmDenoise method has been implemented as an R package, which is freely available on the GitHub repository, https://github.com/YSKoseki/gmmDenoise.