Background:
The secondary structure of RNA can be predicted by the thermodynamics-based method of Zuker and Turner. The accuracy of the method's secondary structure predictions for rRNA can be assessed by using as reference the currently available rRNA secondary structure models that have been derived from comparative analysis of rRNA sequence alignments.

Results:
We folded 72 23S rRNA sequences with the Zuker-Turner method and scored the resulting secondary structure predictions against the comparative model. Empirically, trends in the score were observed as a function of the phylogenetic memberships of the sequences and as a function of the base pairs secondary structural contexts. Further, three parameters were found that (anti-)correlate with the score.

Conclusions:
Three semiquantitative predictors of score were found: % of noncanonical base pairs, % of hairpin loops that were stable tetraloops, and sequence %G+C. The folding of rRNA is a tractable problem and thermodynamics-based folding algorithms, in particular, are useful in the study of this folding problem even for large RNA molecules (e.g. 16S and 23S rRNA).