Our understanding of the role of the RNA has changed considerably in recent decades. Advances in molecular biology have shown that some so-called non-coding RNAs (ncRNAs) play different roles in several stages of the life of the cell. These ncRNAs are transcribed from DNA, but, unlike messenger RNAs, they do not code for a protein and are functional. It is now a major challenge to find these new ncRNA in sequenced genomes, and several approaches can be used. One of these approaches focuses on the localization of new members of known ncRNA family. The approach supposes a prior knowledge of the signature of the family, i.e., the conserved sequence and structural elements of the known members of a family. The aim is to find all the regions in a genome that match the signature. However, none of these softwares is based on any precise formalism, their efficiency is variable, and they often lack clear scoring system to enable ranking of solutions. Moreover, none of them, including Infernal, accepts RNA-RNA inter-molecular interactions, which are required to accurately describe some ncRNA families We first proposed MilPat that makes possible to model RNA-RNA interactions, as well as several other sequence and structural elements. MilPat is based on the constraint network framework, and gives good results on the ncRNA localization problem. However, MilPat does not support costs, which is the major drawback of the approach. In this talk, I will present Darn!, which extends the approach implemented in MilPat, by integrating costs, as well as some other mechanisms.