Benthic diatoms are routinely used as ecological indicators in rivers. A standardized methodology is based on biofilm sampling, species identification, and counting under microscope. DNA-metabarcoding is an alternative methodology that can identify species and assess their proportion based on high-throughput DNA sequencing. Sequence data is analyzed with bioinformatics tools, and several strategies can be chosen. The strategy choice can affect communities composition and structure, and therefore the resulting ecological assessment. We wanted to optimize the bioinformatics strategy to obtain the closest results to microscopy. This was done in the framework of the Mothur pipeline. Here, 447 samples from French rivers were analyzed in the monitoring context of the European Water Framework Directive. Samples were analyzed both with DNA metabarcoding and microscopy. A usual bioinformatics strategy in Mothur includes clustering DNA-sequences into Operational Taxonomic Units (OTUs). Different algorithms exist for this. From a subsample of 142 samples, we showed that some strategies (Furthest neighbor) gave closer results to microscopy than others (Opticlust) in terms of community structure and diatom index values. However, we showed that OTU clustering was not necessary for ecological monitoring: Direct taxonomic assignment of individual sequence units (ISU) gave similar results to those obtained in microscopy. Interestingly, direct assignment enabled the detection of more species 2 to 3 times faster in terms of computation time compared to the OTU strategy. However, it remained important to remove low quality and chimeric sequences; if not, biomonitoring results differed greatly from microscopy. We showed that it was preferable to have a loose taxonomical identification threshold instead of a stringent one. This allowed detecting more species, which could participate in the index calculation and increased its performance. Indeed, in diatoms, phylogenetically neighbor species often have similar ecologies, and this explains why it is preferable, in a biomonitoring framework, to identify more species with less stringency instead of identifying few species with stringency. Finally, the best strategy (direct assignment of filtered ISU with a loose taxonomical threshold of 60%) was applied to the 447 samples covering a large diversity of ecological qualities. These data were then used to produce quality index values, using a quantification correction factor taking into account species biovolumes. Compared to microscopy, the DNA-based method assigned the same quality class for 66% of the samples, and 72% of the samples had an index value (ranging from 0 to 20) with less than one point difference from microscopy.