Terminal restriction fragment length polymorphism (T-RFLP) analysis of amplified ribosomal RNA genes is used for profiling microbial communities and sometimes for species richness and relative abundance estimation in environmental samples. However, the T-RFLP fingerprint may be subject to biases during the procedure, influencing the detection of real community structures in the environment. To investigate possible sources of T-RFLP bias, 18S rRNA gene clones derived from two arbuscular mycorrhizal fungal sequences were combined in simple pairwise mixes to assess the effects of polymerase chain reaction cycle number, plant genomic DNA purification method and varying template ratio on the template-to-product ratio as measured by relative T-RF peak area. Varying cycle numbers indicated that amplification was still in the exponential phase at the cycle numbers lower than 18, so these small cycle numbers were used for the comparison of template-to-product quantities. Relative abundance estimated from T-RF peak ratios varied with different purification procedures, but the best results, closest to input ratios, were obtained by using phenol-chloroform purification. The presence of an excess of unpurified non-target plant genomic DNA generated a bias towards lower or overestimation of relative abundance. We conclude that a low number of amplification cycles and stringent DNA purification are necessary for accurate mixed sample analysis by T-RFLP.