We present a strategy for phylogenetic marker development in non-model systems. Rather than using the traditional approach of comparing distantly related taxa to develop conserved primers for unknown species, we explore an alternative strategy that builds primers directly from a single, relatively well characterized species and applies those primers to increasingly distantly related taxa. We develop and test our protocol with turtles. Using a single BAC end-sequence library consisting of 3461 sequences totaling 2.43 million base pairs of data, we outline a procedure to flag repeat elements, followed by a BLAST approach to categorize sequences into high, low, and no similarity compartments compared to GenBank sequences. We developed and tested a panel of 96 primer pairs with a set of turtle tissues that forms a series of increasingly distantly related taxa with respect to the BAC reference species. Finally, we sequenced 11 of these newly discovered markers across a diverse set of 18 turtle species that spans the 210 million years of chelonian crown-group history and that includes representatives of most of the major clades of extant turtles. Our results indicate that large numbers of new, phylogenetically informative markers can be developed quickly and inexpensively from a single BAC, EST, or similar genomic resource, and that those markers provide reliable phylogenetic information across both shallow and deep levels of phylogenetic history. Our results also highlight the importance of screening for and managing repetitive elements found in randomly sequenced DNA fragments. We presume that our strategy should work well across any similarly divergent clade, suggesting that many-marker datasets can be developed quickly and efficiently for phylogenetic analysis.