An approach is described for obtaining 'perfect probes' for type I modular polyketide synthase (PKS) gene clusters that in turn enables the identification of all such gene clusters in a genome. The approach involves sequencing small fragments of a random genomic DNA library containing one or more modular PKS gene clusters, and identifying which fragments emanate from PKS genes. Knowing the approximate sizes of the genome and the target gene cluster, one can predict the the frequency that a PKS gene fragment will be present in the library sequenced. Computer simulations of the approach were applied to the known PKS and non-ribosomal peptide synthetase (NRPS) gene clusters in the Bacillus subtilus genome. The approach was then used to identify PKS gene fragments in a strain of Sorangium cellulosum that produces epothilone. In addition to identifying fragments of the epothilone gene cluster, we obtained 11 unique fragments from other PKS gene clusters; the results suggest that there may be six to eight PKS gene clusters in this organism. In addition, we identified four unique fragments of NRPS genes, demonstrating that the approach is also applicable for identification of these modular gene clusters.