The viral erb A oncogene is a mutated allele of a normal cell gene for a thyroid hormone receptor. The DNA recognition properties of the v-erb A protein are altered from those of the thyroid hormone receptor, due in part to a point mutation in the P-box of the zinc-finger domain of the viral allele. We report here the effects of systematically varying this P-box codon; our results suggest that this P-box amino acid contributes to DNA specificity not by promoting recognition of the appropriate response elements, but rather by excluding binding of the erb A protein to inappropriate half-sites. In this manner, DNA recognition by the v-erb A protein appears to differ from that by the glucocorticoid receptor. A variety of P-box amino acids were compatible with recognition of the prototypic AGGTCA half-site; intriguingly, several of these mutant erb A proteins could also recognize a variety of alternative half-site sequences. Recognition of these alternative half-sites required a compatible amino acid sequence in the N terminus of the erb A protein. Our results begin to define a code by which the identity of the amino acids in the zinc-finger and N-terminal domains is reflected in the DNA recognition properties of the receptor.