Runs of adenines (adenine tracts) have been implicated as the main determinant of sequence-directed DNA bending. The most widely used experimental test for bending relies on the observation that bent DNA migrates more slowly than straight DNA on a polyacrylamide electrophoresis gel. It was shown recently that the polymer (GTTTTAAAAC)n runs with normal mobility on a gel, whereas (GAAAATTTTC)n runs more slowly and thus appears to be strongly bent. The observation that these similar sequences, which differ only in the order of the adenine and thymine tracts, adopt such different shapes offers a stringent test of theories to explain DNA bending. Although the wedge model for DNA bending has recently been elaborated to explain the gel mobilities of these molecules, we wished to determine experimentally the structural basis for the difference in bending. We report here measurements of the frequency of cleavage by the hydroxyl radical at each nucleotide of cloned versions of the two polymers (see Fig. 1). We show that the TTTTAAAA sequence does not display the cleavage pattern that is associated with bent DNA, whereas the AAAATTTT sequence does. The observed sequence dependence of the cleavage pattern of an adenine tract is at odds with current models for DNA bending, which assume that adenine tracts always adopt the same conformation.