This study has two main objectives: (1) to implement a recycling aquaculture system (RS) for the larvae of the oyster Crassostrea gigas, and (2) to characterise the bacterial communities established in different compartments of this system. An RS with 25% fresh seawater addition per hour and another with no addition (0%) were compared with a flow-through system (FT). Larval survival was equivalent in RS and FT, but growth rate was 17% slower in RS than in FT. The physical chemical parameters remained stable, except for pH that decreased to 7.75 and salinity that increased to 37.5 ‰ in the RS 0%. In both systems, the cultivable bacteria were present in equal numbers in seawater (around 105 ml-1) and in larvae (103 larva-1) on day 15. Bacterial assemblages, characterised by 454 pyrosequencing of the V1–V3 region of 16S rRNA, were highly similar (50–65%) for compartments, regardless of rearing system and sampling time, but the compartments were clearly different from one another. At the beginning of rearing, larval microbiota was mostly composed of Proteobacteria (~90%), with 47% Rhodobacteraceae (a-Proteobacteria). ?-Proteobacteria, including Pseudoalteromonas, Alteromonas and a few vibrios, declined in the rearing period (25% on day 7 to 9% on day 15). At the end of rearing, colonisation by two members of the Burkholderiales (ß-Proteobacteria), 45% on average on day 15, had decreased overall diversity. Seawater microbiota was more stable with in all batches (27 ± 7%), one unclassified bacterium present 42 member of a-Proteobacteria (19 ± 7%) and 26 of ?-Proteobacteria (15 ± 0%). Change was due notably to a species of Cryomorphaceae (Flavobacteria) that reached 15 ± 7% on day 15. Predatory bacteria, Bdellovivrio spp. and Bacteriovorax spp. were present (3–12%) and could participate in the regulation of bacterial populations. Bacterial assemblages in RS bioreactors remained stable and were mainly composed of Rhodobacteraceae, Rhizobiales and Planctomycetes. Only a few nitrite oxidisers were detected and no ammonia oxidisers, although nitrification was efficient in RS without addition of new water. The larval microbiota was made up of bacteria growing in seawater, but some such as the Burkholderiales could have come from the broodstock. Several bacteria predominant in seawater were first harboured in the algal culture. Finally, despite sanitary measures in the hatchery (UV treatment and frequent cleaning), the diversity of microbiota remained high, although it did not contain pathogenic bacteria. The presence of this microbiota might even be indispensable to ensure normal ontogenesis, as suggested by the holobiont concept.