Total DNA from sediment samples was isolated by a direct lysis technique. Purified DNA was used as template either undiluted or diluted 1:10 prior to polymerase chain reaction (PCR) amplification of 16S rRNA genes. Full-length inserts were analysed for restriction fragment length polymorphisms (RFLP) with the enzyme Cfo1, and the resulting distribution and abundance of RFLP patterns compared between the undiluted and diluted PCR reactions. Results indicate that for low PCR template concentrations, in the range from a few picograms to tens of picograms DNA, proportional representation of specific RFLP types was not reproducible upon template dilution, confirming that PCR amplification of 16S rDNA cannot be used directly to infer microbial abundance. In particular, only 15-24% of the RFLP types recovered from a sample were present in both the undiluted and diluted extracts. We propose that very low template concentrations in the PCR generate random fluctuations in priming efficiency, which led to the contrast in RFLP types observed in the libraries from the undiluted and diluted extracts.