High-throughout sequencing platforms are widely used in metabarcoding studies of environmental microbial diversity as they can quickly produce millions of reads. Resulting reads in these studies are clustered into molecular operational taxonomic units (OTUs) and compared statistically. In order to limit the number of spurious OTUs retrieved from samples, eliminating the reads with mutated primers has become the norm. This process has the advantage of not requiring the use of complex tools, since it is possible to search for exact primers with simple regular expressions natively supported by many programming languages (python, perl, ruby, etc.). However, this strategy may also eliminate correct sequences and this practice raises questions: Does the removal of all reads with mutated primers cause information loss? Or, in a more practical perspective : now there are tools to reject the less reliable sequences in clustering as SWARM [cite:mahe_swarm_2015], is there an interest to seek mutated primers? Can it bring new sequences? Are these sequences relevant? Can it contribute to detect more species? Such are the questions that we try to answer in this article, through a metagenomic analysis that estimates eukaryotic soil biodiversity. To try to answer these questions, we analysed data on tropical soils to study the impact on metabarcoding results of keeping reads with mutated primers. Main results: The study shows that keeping such reads allows identifying more sequences. The majority of the new sequences are quite similar to sequences with exact primers. A minority of the new sequences contributes to validate new clusters as potential species signatures.