Advances in DNA metabarcoding have greatly expanded our knowledge of microbial communities in recent years. Pipelines and parameters have been tested extensively for bacterial metabarcoding using the 16S rRNA gene and best practices are largely established. For fungal metabarcoding using the ITS gene, however, only a few studies have considered how such pipelines and parameters can affect community prediction. Here we report a novel bias uncovered during ITS2 sequencing ofTrichoderma-infected ant fungus gardens and confirmed using mock communities. Abnormally low forward read quality causedTrichodermaITS2 reads to be computationally filtered before and during read pair merging, thus almost entirely eliminatingTrichodermaASVs from the resulting fungal community profiles. Sliding window quality trimming before filtering allowed most of these reads to pass filtering and merge successfully, producing community profiles that now correlated with visual signs ofTrichodermainfection and matched the composition of the mock communities. Applying such sliding window trimming to a previously generated environmental ITS2 dataset increased the detected fungal diversity and again overcame read quality biases againstTrichodermato instead detect it in nearly every sample and often at high relative abundances. This analysis additionally identified a similar, but distinct, bias against a second fungal genusMeyerozyma. The prevalence of such quality biases against other fungal ITS sequences is unknown but may be widespread. We therefore advocate for routine use of sliding window quality trimming as a best practice in ITS2 metabarcoding analysis.ImportanceMetabarcode sequencing produces DNA abundance profiles that are presumed to reflect the actual microbial composition of the samples that they analyze. However, this assumption is not always tested, and taxon-specific biases are often not apparent, especially for low-abundance taxa in complex communities. Here we identified ITS2 read quality aberrations that caused dramatic reductions in the relative abundances of specific taxa in multiple datasets characterizing ant fungus gardens. Such taxon-specific biases in read quality may be widespread in other environments and for other fungal taxa, thereby causing incorrect descriptions of these mycobiomes.