Similarity searching is an important tool to many biological scientists. Various computer implementations (BLAST, FASTA, Smith-Waterman) are used by scientists to analyze their sequences of interest to identify identities (perfect matches) or similarities (statistically significant matches) between their query sequences and large databases such as GenBank. Search engines currently return brief annotations and alignments ranked in order of statistical significance or raw similarity score. However, it is frequently not the top-scoring similarities that bring important new information to the investigating scientist, but the content of the annotation or similarity hits at any significant score. The Gene Alert algorithm applies additional filtering and a user weighted keyword search to the BLAST output to parse the output into a form customized to the user. There are three components to the Gene Alert implementation as it is currently operating: an organized file structure, a BLAST engine, and a parser written in the PERL scripting language. The file structure was designed to place code and database components in logical positions and to facilitate future complete automation of the Gene Alert and similarity search system. Shown here is the file structure within the UNIX environment.