Three soil DNA extraction procedures (homemade protocols and commercial kit) varying in their practicability were applied to contrasting soils to evaluate their efficiency in recovering: (i) soil DNA and (ii) bacterial diversity estimated by 16S rDNA pyrosequencing. Significant differences in DNA yield were systematically observed between tested procedures. For certain soils, 10 times more DNA was recovered with one protocol than with the others. About 15 000 sequences of 16S rDNA were obtained for each sample which were clustered to draw rarefaction curves. These curves, as well as the PCA ordination of community composition based on OTU clustering, did not reveal any significant difference between procedures. Nevertheless, significant differences between procedures were highlighted by the taxonomic identification of sequences obtained at the phylum to genus levels. Depending on the soil, differences in the number of genera detected ranged from 1% to 26% between the most and least efficient procedures, mainly due to a poorer capacity to recover populations belonging to Actinobacteria, Firmicutes or Crenarchaeota. This study enabled us to rank the relative efficiencies of protocols for their recovery of soil molecular microbial biomass and bacterial diversity and to help choosing an appropriate soil DNA extraction procedure adapted to novel sequencing technologies.