Desertomycin A is an aminopolyol polyketide containing a macrolactone ring. We have proposed that desertomycin A and similar compounds (marginolactones) are formed by polyketide synthases primed not with gamma-aminobutanoyl-CoA but with 4-guanidinylbutanoyl-CoA, to avoid facile cyclization of the starter unit. This hypothesis requires that there be a final-stage de-amidination of the corresponding guanidino-substituted natural product, but no enzyme for such a process has been described. We have now identified candidate amidinohydrolase genes within the desertomycin and primycin clusters. Deletion of the putative desertomycin amidinohydrolase gene dstH in Streptomyces macronensis led to the accumulation of desertomycin B, the guanidino form of the antibiotic. Also, purified DstH efficiently catalyzed the in vitro conversion of desertomycin B into the A form. Hence this amidinohydrolase furnishes the missing link in this proposed naturally evolved example of protective-group chemistry.