Rate variation of evolution among-site in rRNA genes is a potential problem leading to a systematic error in phylogenetic inference. In this study, we applied different partitioning strategies based on the characters of rRNA secondary structure to examine this problem. Based on a phylogenetic analysis of 52 fungal taxa, we assessed the performance of different partitioning schemes on Bayesian inference. The comparison of evolutionary models demonstrated the relative sensitivities of the best-fit models and parameter estimates to different structural partitions. In contrast to traditional unpartitioned method, partitioning schemes based on loop elements of secondary structure have little effects on phylogenetic analyses, while the use of stem elements improve the marginal likelihoods and the ability to estimate phylogenies. Additionally, despite strong support by Bayes factors, simply including more partition subsets do not to improve ability to estimate phylogenies, which means that biological factors (or secondary structure characters) instead of mathematics ones should be considered to yield a reasonable partitioning strategy for rRNA genes.