The single-crystal x-ray analysis of orthorhombic CATGGCCATG has revealed a previously unrecognized mode of intrinsic bending in DNA. The decamer shows a smooth bend of 23 degrees over the central four base pairs, caused by preferential stacking interactions at guanine bases. The bend is produced by a roll of base pairs along their long axes, in a direction that compresses the wide major groove of the double helix. This major-groove-compressing bend at GGC, plus the abundant crystallographic evidence that runs of successive adenine bases (A-tracts) are straight and unbent, requires rethinking of the models most commonly invoked to explain A-tract bending. A decade of excellent experimental work involving gel migration experiments, cyclization kinetics, and nucleosome phasing has clearly established that introduction of short A-tracts into a general DNA sequence in synchrony with the natural repeat of the helix leads to bending. But it does not logically and inevitably follow that the actual bending is to be found within these introduced A-tracts or even at junctions with general-sequence B-DNA.