Reliability of bacterial diversity assessment using polymerase chain reaction (PCR) denaturing gradient gel electrophoresis (DGGE) analysis of 16S rDNA fragments was evaluated for a particular complex microbial assemblage: river epilithic biofilm. By comparing 3 routine protocols on replicates of one river biofilm sample, we found that common DNA extraction procedures gave comparable diversity (from 28.0 to 30.7 bands detected) and community composition (> 75% of homology) despite differences in the total amount of extracted DNA (from 0.9 to 4.2 microg). Therefore methodological improvements only concerned electrophoretic separation of DNA fragments (range of denaturing gradient from 35% to 70% and migration time=18h) and standardisation of DNA amounts used (PCR-template=50 ng, gel loading=700 ng). Using such a standardised methodology we found a good reproducibility of all steps of the procedure. When an Escherichia coli strain was introduced as a contaminant in a biofilm sample, we were able to recover ribotypes from the strain. As concerns fields sampling, a satisfactory repeatability of banding patterns from neighbouring pebbles (sampling point) allowed discriminating between the biofilm intrasite variability (various points from a cross-profile). These trials confirmed that PCR-DGGE is suitable to assess a reliable genetic fingerprint of epilithic biofilms in the river. Phylogenetic analysis of 40 partial sequences of 16S rDNA from DGGE gels of two sets of river biofilms samples proved evidences for the retrieval of DNA fragments related to phototroph Eukarya. However, in both cases plastidial 16S rDNA represented less than 25% of the analysed operational taxonomic units. Taking into account that Cyanobacteria, as members of the Bacteria, were also detected, sequence analysis of relevant bands from the pattern is required to target "bacteria", i.e. the functional group of prokaryotic microorganisms to which one commonly refers as a key component in sustaining the nutrient turnover.