Experiments with d(TA)N§ § Abbreviations used: The conventions of Inman & Baldwin (1962) are followed. hairpin helices are reported which contradict the simple Ising (or nearest-neighbor interactions) model for the DNA helix-coil transition at low or moderate counterion concentrations. The experiments show that electrostatic interactions which extend beyond nearest neighbors must be taken into account. The basic observations are as follows. First the dependence of Tm(N) on log M (where M is monovalent counterion molarity) falls off sharply with decreasing N in the range 8 ≤ N ≤ 22, so that the set of melting curves is compressed into a narrow temperature range at low M. At M = 0.0028, d(TA)20 has a higher Tm than poly d(AT) and the Tm of d(TA)9 is only slightly lower. Second, electrostatic interactions have a relatively small effect on the breadths of these melting curves: the curves for all oligomers sharpen slightly as the counterion concentration is reduced. We have used the method of Schildkraut & Lifson (1965) to study the problem by a priori computation. Their method is based on a summation of discrete pairwise interactions between the charges on the phosphate groups. As in the theory of Record (1967), the quantity zp2 D is taken to be the sole adjustable parameter, where zp is the effective charge per phosphate group and D is the effective dielectric constant. Two results are found: (1) the melting curves for all oligomers can be represented for all values of M by a single value of zp2 D, and (2) it is necessary to count interactions between charges both in the helix and in adjoining non-helical segments. There is a simple approximation by which electrostatic effects may be taken into account while retaining much of the formalism of standard helix-coil transition theory. In this approximation the stability constant s is replaced by an effective constant s*(N), which is a function of 2N, the number of nucleotides in an oligomer, although independent of helix length in a partly helical molecule. The melting curves can be predicted in a straightforward manner using this approximation and they agree with the experimental curves. However, further work is needed to define the limitations of this approximation.