The performance and microbial communities of three differently inoculated biotrickling filters removing dimethyl sulfide (DMS) were compared. The biotrickling filters were inoculated with Thiobacillus thioparus TK-m (TRIO), sludge (HANDS) and sludge + T. thioparus TK-m+Hyphomicrobium VS (HANDS++), respectively. The criteria investigated were length of the start-up period, the maximum elimination capacity, and the effects of intermittent loading rates, low pH, peak loading and very low loading rate on the DMS removal efficiency. The HANDS++ reactor exhibited the best performance considering all treatments. HANDS performed almost equally well as HANDS++, except during the determination of the ECmax, while THIO was generally the least efficient. During stable DMS loading at concentrations of 20 ppmv or lower, all reactors exhibited similar and high removal efficiencies (> 99%). Denaturing gradient gel electrophoresis (DGGE) analysis showed the establishment of T. thioparus in the biofilm of all reactors, but not of Hyphomicrobium VS. Quantitative monitoring of the introduced bacterial strains was performed with a newly developed real-time PCR protocol. Initially, the inoculated strains were exclusively found in the reactors in which they were added. Afterwards, however, both strains developed in the biofilm of all three reactors, although T. thioparus attained higher cell densities than Hyphomicrobium. The presence of T. thioparus in TRIO was related with the DMS loading rates that were applied, in the sense that intermittent DMS loading and very low DMS loading rates (0.5 ppmv) induced a decrease in gene copy numbers. Real-time PCR and DGGE both gave consistent results regarding the presence of Hyphomicrobium VS and Thiobacillus thioparus TK-m in the reactors. Only real-time PCR could be used to detect bacteria comprising of less than 1.4% of the total bacterial community (similar to 10(5) copies ring(-1)). (c) 2006 Wiley Periodicals, Inc.