Our study of 164 diatom samples from Catalonia (NE Spain) is the first to evaluate the applicability of DNA metabarcoding, based on high throughput sequencing (HTS) using a 312-bp rbcL marker, for biomonitoring Mediterranean rivers. For this, we compared the values of a biotic index (IPS) and the ecological status classes derived from them, between light microscope-based (LM) and HTS methods. Very good correspondence between methods gives encouraging results concerning the applicability of DNA metabarcoding for Catalan rivers for the EU Water Framework Directive (WFD). However, in 10 sites, the ecological status class was downgraded from “Good”/“High” obtained by LM to “Moderate”/“Poor”/“Bad” by HTS; these “critical” sites are especially important, because the WFD requires remedial action by water managers for any river with Moderate or lower status. We investigated the contribution of each species to the IPS using a “leave-one-out” sensitivity analysis, paying special attention to critical sites. Discrepancies in IPS between LM and HTS were mainly due to the misidentification and overlooking in LM of a few species, which were better recovered by HTS. This bias was particularly important in the case of Fistulifera saprophila, whose clear underrepresentation in LM was important for explaining 8 out of the 10 critical sites and probably reflected destruction of weakly-silicified frustules during sample preparation. Differences between species in the rbcL copy number per cell affected the relative abundance obtained by HTS for Achnanthidium minutissimum, Nitzschia inconspicua and Ulnaria ulna, which were also identified by the sensitivity analysis as important for the WFD. Only minor IPS discrepancies were attributed to the incompleteness of the reference library, as most of the abundant and influential species (to the IPS) were well represented there. Finally, we propose that leave-one-out analysis is a good method for identifying priority species for isolation and barcoding.