Fungi play critical roles in a variety of ecosystems, including forests, driving nutrient cycling or maintaining ecological stability. Despite this importance, capturing the broad extent of fungal diversity and its temporal dynamics remains constrained by methodological limitations. By using environmental DNA (eDNA) metabarcoding complemented by culture-based analyses, this study investigates fungal community changes within leaf litter across three dominant tree species (Fagus japonica, F. crenata, and Quercus serrata) in the Ogawa Forest Reserve, Japan, over a 30-year period. We analyzed samples from seven time points (1992, 1993, 1994, 1995, 2003, 2013, and 2023) to track the temporal dynamics of fungal assemblage. Our eDNA metabarcoding approach detected 265 operational taxonomic units (OTUs), with notably higher species richness in recent samples (226 OTUs in 2023 versus 25 OTUs in 1992). For the specimens in 2023, a comparison between eDNA from litter traps and cultures from freshly collected forest samples revealed notable methodological differences. The eDNA method identified 200 unique OTUs compared to only 25 OTUs detected via culture-based methods, highlighting the superior sensitivity of molecular approaches for biodiversity assessment. Temporal analysis revealed distinct community composition patterns along the three-decade gradient. The intermediate sampling years demonstrated a gradual transition in dominant taxa, potentially driven by environmental changes and succession dynamics within the forest ecosystem or by stochastic drift. Host specificity emerged as a significant factor shaping fungal communities. Each tree species harbored distinct fungal assemblages, with Q. serrata supporting the most diverse fungal communities across half of the sampling time points (except in 1992, 1993, and 1995). Non-metric multidimensional scaling (NMDS) analysis confirmed clear clustering of fungal communities by both tree species and sampling year, indicating both spatial and temporal structuring of fungal diversity. While potential contamination during long-term sample storage cannot be entirely ruled out, the observed host-specificity patterns suggest our detected temporal patterns reflect genuine ecological shifts rather than storage artifacts. The detection of identifiable fungal DNA from 30-year-old samples represents a significant methodological achievement, validating the reliability of our approach for reconstructing historical fungal communities. Future work will explore functional implications of these community transitions on forest ecosystem processes.