DNA metabarcoding of freshwater communities typically relies on PCR amplification of a fragment of the mitochondrial cytochrome c oxidase I (COI) gene with degenerate primers. The advantage of COI is its taxonomic resolution and the availability of an extensive reference database. However, when universal primers are used on environmental DNA (eDNA) isolated from water, benthic invertebrate read and OTU numbers are typically “watered down,” that is, under represented, compared to whole specimen “bulk samples” due to greater co-amplification of abundant nontarget taxa (e.g., fungi, algae, and bacteria). Because benthic stream invertebrate taxa are of prime importance for regulatory biomonitoring, more effective ways to capture their diversity via eDNA isolated from water are important. In this study, we aimed to improve benthic invertebrate assessment from eDNA by minimizing nontarget amplification. Therefore, we generated eDNA data using universal primers BF2/BR2 on samples collected throughout 15 months from a German Long-Term Ecological Research site (Rhine-Main-Observatory, Kinzig River) to identify most abundant nontarget taxa. Based on these data, we designed a new reverse primer (EPTDr2n) with 3’-specificity toward benthic invertebrate taxa and validated its specificity in silico together with universal forward primer fwhF2 using available data from GenBank and BOLD. We then performed in situ tests using 20 Kinzig River eDNA samples. We found that the percentage of target reads was much higher for the new primer combination compared to two universal benthic invertebrate primer pairs, BF2/BR2 and fwhF2/fwhR2n (99.6% versus 25.89% and 39.04%, respectively). Likewise, the number of detected benthic invertebrate species was substantially higher (305 versus 113 and 185) and exceeded the number of 153 species identified by expert taxonomists at nearby sites across two decades of sampling. While few taxa, such as flatworms, were not detected, we show that the optimized primer avoids the nontarget amplification bias and thus significantly improves benthic invertebrate detection from eDNA.