Reliable abundance estimation is a primary challenge in environmental DNA (eDNA) analysis, which has been addressed by considering the effects of eDNA transport and degradation. However, these eDNA spatial dynamics depend on the cellular and molecular structure of eDNA, with its persistence state (particle size and DNA fragment length) being essential for improved abundance estimation. This existing knowledge gap is bridged by utilizing datasets obtained from two types of aquarium experiments (targeting zebrafish [Danio rerio] and Japanese jack mackerel [Trachurus japonicus]) and comparing the relationships between eDNA concentration and species abundance among different eDNA size fractions and target marker lengths. We reared the fish in experimental tanks with different individual numbers or biomass densities, filtered rearing water using different pore size filters, and quantified eDNA concentrations targeting different fragment lengths or genetic regions. Consequently, both experiments showed that the accuracy and sensitivity in abundance estimation were improved (i.e., R² values and slopes of linear regressions increased) when targeting eDNA at the 3–10-µm size fraction. On the other hand, targeting eDNA at the >10 µm size fraction yielded a lower R² value. This result indicates that an “appropriately” larger eDNA particle is vital for improving abundance estimation accuracy and sensitivity. Conversely, the target marker length negatively affected the R² value. This study proposes that the relationship between eDNA concentration and species abundance relies on the complex interactions between the particle size, persistence, and spatial heterogeneity of eDNA in water.