The predominance of rRNAs in the transcriptome is a major technical challenge in sequence-based analysis of cDNAs from microbial isolates and communities. Several approaches have been applied to deplete rRNAs from (meta)transcriptomes, but no systematic investigation of potential biases introduced by any of these approaches has been reported. Here we validated the effectiveness and fidelity of the two most commonly used approaches, subtractive hybridization and exonuclease digestion, as well as combinations of these treatments, on two synthetic five-microorganism metatranscriptomes using massively parallel sequencing. We found that the effectiveness of rRNA removal was a function of community composition and RNA integrity for these treatments. Subtractive hybridization alone introduced the least bias in relative transcript abundance, whereas exonuclease and in particular combined treatments greatly compromised mRNA abundance fidelity. Illumina sequencing itself also can compromise quantitative data analysis by introducing a G+C bias between runs.