Until recently, the de-facto method for short read-based amplicons reconstruction is a sequence similarity threshold approach (Operational taxonomic Units OTUs). This assumption was relaxed by shifting to Amplicon Sequencing Variants (ASVs) where distributions are fitted to abundance profiles of individual genes using a noise-error model. Whilst OTUs-based approach is still useful for 16SrRNA/18S rRNA regions, where typically 97-99% thresholds are used, their utility to functional genes is still debatable as there is no consensus on how to cluster the sequences together. Here, we compare OTUs- and ASVs-based reconstruction approaches as well as taxonomy assignment methods, Naïve Bayesian Classifier (NBC) and Bayesian Lowest Common Ancestor Algorithm (BLCA), using functional genes dataset from the microbial nitrogen-cycling community in the Brouage mudflat (France). A range of OTU similarity thresholds and ASV were used to compare amoA (AOA and AOB), nxrB, nirS, nirK and nrfA communities between differing sedimentary structures. We show that for AOA-amoA and nrfA, the use of ASV led to differences in the communities between sedimentary structures whereas the use of OTUs didn’t. Conversely, significant differences were detected when using OTU (97%) for AOB-amoA but not with ASV or OTUs at other similarity thresholds. Interestingly, conclusions drawn from the other three functional genes were consistent between amplicon reconstruction methods. We also show that, when the sequences in the reference-database are related to the environment in question, BLCA leads to more phylogenetically relevant classifications. However, when the reference database contains sequences more dissimilar to the ones retrieved, NBC helps obtain more information.