The twisting potential of DNA has been determined directly by a method that measures the cyclization probability or j-factor of EcoRI restriction fragments as a function of DNA twist. The cyclization probability is proportional to Kc, the equilibrium constant for cyclization of the restriction fragment via its cohesive ends (Shore et al., 1981). Here we vary the twist of the DNA by making small internal additions to or deletions from a 242 bp EcoRI restriction fragment. A series of 12 DNA molecules has been studied, which range in length from 237 to 254 bp. The cyclization probability is measured from the rates of covalent closure by phage T4 DNA ligase of two systems: (1) a linear restriction fragment in equilibrium with its cyclized form and (2) half molecules (cut by a blunt-end endonuclease) in equilibrium with joined half molecules. The striking result is that, in this DNA size range, the j-factor depends strongly on the fractional twist: the difference between the total helical twist and the nearest integer. Thus j depends in an oscillatory manner on DNA length between 237 and 254 bp with a period of about 10 bp. These data give the free energy of DNA twisting as a function of twist. The curve of j versus DNA length can be fitted to a harmonic twisting potential with a torsional constant of C = 2·4 × 10−19 erg cm. This value is in reasonable agreement with different estimates of C made by Barkley & Zimm (1979: C = 1·8 × 10−19 to 4·1 × 10−19 erg cm) and is somewhat larger than the value obtained resulting from the kinetics of DNA twisting measured by fluorescence depolarization of ethidium intercalated into DNA (C = 1·4 × 10−19 erg cm; Millar et al., 1982; Thomas et al., 1980) or from spin label studies (Hurley et al., 1982). Our experiments provide a direct measurement of the torsional free energy and they show that the DNA twisting potential is symmetric. Our experiments also indicate that the DNA helix is continuous, or nearly so, in a nicked circle; presumably this happens because the DNA stacking interaction maintains the double helix in register across a single-strand nick. As a consequence, the twist of a singly nicked DNA circle is integral for small (≊250 bp) planar DNA circles and there is a change in twist upon cyclization. This conclusion is supported by finding that the rate of closure by T4 ligase of singly nicked circles is not dependent on either small or large changes in DNA length, in contrast to the results given above for cyclization of linear restriction fragments. The temperature dependence of the ligase closure reaction has also been measured for three systems: (1) linear restriction fragments in equilibrium with cyclized molecules; (2) singly nicked circles; and (3) half molecules in equilibrium with joined half molecules. The temperature dependence data confirm that systems (1) and (3) are in equilibrium as measured by the rate of the ligase closure reaction: the rates of these two reactions increase strongly with decreasing temperature, in response to the increasing fraction of joined cohesive ends at low temperatures, whereas the rate of closing singly nicked circles increases with increasing temperature, as expected for a typical enzyme-catalyzed reaction. The ratio of rate constants measured for (1) and (2) gives Kc, the equilibrium constant for cyclization. Values obtained in this way show reasonable agreement with electron microscopic measurement of Kc by Mertz & Davis (1972). The ratio of rate constants measured for (3) and (2) gives Ka, the equilibrium constant for bimolecular association of half molecules, and the temperature dependence of Ka gives a reasonable value (−32(±10) kcal/mol) for the enthalpy of joining the EcoRI cohesive ends.