The confident discrimination of nucleic acids that share a high degree of sequence identity is the major obstacle for the widespread applicability of multiplex DNA-based techniques. This diagnostic uncertainty originates in the insufficient specificity of hybridization, allowing cross-hybridization between unwanted probe-target combinations. Starting from a random mixture of oligonucleotides, we describe a protocol to selectively amplify the probes that bind to the target but not to the similar, unintended targets. The procedure involves five forward hybridizations to generate pools of probes with significant affinity, but not necessarily specificity, for the target. Specificity is then achieved during subtractive hybridization steps, where only probes having differential diagnostic performance are retained. Iterative hybridizations, cloning, sequencing and testing of the performance of selected probes can all be fully automated. Eight weeks are required for the full completion of a project composed of 40 probe-target pairs, even when targets share as much as 87% of sequence identity. While alternative, computer-assisted, rational oligonucleotide design may produce an uncertain outcome, the present protocol generates robust and specific probes suitable for a variety of multiplex, nucleic acid-based detection/typing platforms.