Given an RNA family characterized by conserved sequences and folding constraints, the problem is to search for all the instances of the RNA family in a genomic database. As seed-based heuristics have been proved very efficient to accelerate the classical homology based search methods such as BLAST, we use a similar idea for RNA structures. We present an exclusion method for RNA search allowing for possible nucleotide insertion, deletion and substitution. It is based on a partition of the RNA stem-loops into consecutive seeds and a preprocessing of the target database. This algorithm can be used to improve time efficiency of current methods, and is guaranteed to find all occurrences that contain at least one exact seed.