Bending of 15 to 24 degrees is observed within crystal structures of B-DNA duplexes, is strongly sequence-dependent, and exhibits no correlation with the concentration of MPD (2-methyl-2,4-pentanediol) in the crystallizing solution. Two types of bends are observed: facultative bends or flexible hinges at junctions between regions of G.C and A.T base-pairs, and a persistent and almost obligatory bend at the center of the sequence R-G-C-Y. Only A-tracts are characteristically straight and unbent in every crystal structure examined to date. A detailed examination of normal vector plots for individual strands of a double helix provides an explanation, in terms of the stacking properties of guanine and adenine bases. The effect of high MPD concentrations, in both solution and crystal, is to decrease local bending somewhat without removing it altogether. MPD gel retardation experiments provide no basis for choosing among the three models that seek to explain macroscopic curvature of DNA by means of microscopic bending: junction being, bent A-tracts, or bent general -sequence DNA. Crystallographic data on the straightness of A-tracts, the bendability of non-A sequences, and the identity of inclination angles in A-tract and non-A-tracts B-DNA support only the general-sequence bending model. The pre-melting transition observed in A-tract DNA probably represents a relaxation of stiff adenine stacks to a flexible conformation more typical of general-sequence DNA.