Although more than 30 Escherichia coli promoters utilize the RNA polymerase holoenzyme containing sigmaS (EsigmaS), and it is known that there is some overlap between the promoters recognized by EsigmaS and by the major E. coli holoenzyme (Esigma70), the sequence elements responsible for promoter recognition by EsigmaS are not well understood. To define the DNA sequences recognized best by EsigmaS in vitro, we started with random DNA and enriched for EsigmaS promoter sequences by multiple cycles of binding and selection. Surprisingly, the sequences selected by EsigmaS contained the known consensus elements (-10 and -35 hexamers) for recognition by Esigma70. Using genetic and biochemical approaches, we show that EsigmaS and Esigma70 do not achieve specificity through 'best fit' to different consensus promoter hexamers, the way that other forms of holoenzyme limit transcription to discrete sets of promoters. Rather, we suggest that EsigmaS-specific promoters have sequences that differ significantly from the consensus in at least one of the recognition hexamers, and that promoter discrimination against Esigma70 is achieved, at least in part, by the two enzymes tolerating different deviations from consensus. DNA recognition by EsigmaS versus Esigma70 thus presents an alternative solution to the problem of promoter selectivity.