Peptide mass fingerprinting and database searching with tandem mass spectrometry are two methods commonly employed to identify proteins in a sample. However, up to 90% of peptides can remain unidentified. In this paper, a search-space filter using amino acids identified by a novel de novo methodology is presented. This provides a high-accuracy set of amino acid predictions through exploiting the internal fragmentation of amino acid chains during tandem mass spectrometry. These predictions are used to reduce the number of peptides considered from a non-redundant peptide database. The presence of one confirmed amino acid can be used to reduce the search-database size by between 33% (Leucine) and 83% (Tryptophan). One or more accurate amino acid identifications are made in 18% of simulated and 9% of experimental peptide spectra considered. Given the large proportion of currently unidentified peptides, this method represents a useful tool for increasing peptide identification rates.