Spectral libraries provide a sensitive and accurate method for identifying peptides from tandem mass spectra, complementary to searching genome-derived databases or sequencing de novo. Their application requires comprehensive libraries including peptides from low-abundant proteins. Here we describe a method for constructing such libraries using biological differentiation to "fractionate" the proteome by harvesting adult organs and tissues, and build comprehensive libraries for identifying proteins in zebrafish (Danio rerio) embryos and larvae (an important and widely used model system). Hierarchical clustering using direct comparison of spectra was used to prioritize organ selection. The resulting and publicly available library covers 14,164 proteins, significantly improved the number peptide-spectrum matches in zebrafish developmental stages and can be used on data from different instruments and laboratories. The library contains information on tissue and organ expression of these proteins and is also applicable for adult experiments. The approach itself is not limited to zebrafish, but would work for any model system.