Non-coding RNA (ncRNA) molecules do not code for proteins, but play important regulatory roles in cellular machinery. Recently, different computational methods have been proposed to identify and classify ncRNAs. In this work, we propose a covariance model with multiple Dirichlet mixture priors to identify ncRNAs. We introduce a tool, named Regene, to derive these priors automatically from known ncRNAs families included in Rfam. Results from experiments with 14 families improved sensitivity and specificity with respect to single component priors.