An experimental study has been made of the factors determining the co-operativity of DNA melting, using two types of d(TA) oligomers (linear and circular) with the alternating base sequence... ATAT... The linear oligomers form open hairpin helices which melt chiefly from the open end. The circular oligomers form closed hairpin helices with a loop at each end and melt by enlarging the loops. Analysis of the melting curves for open hairpins (in 0.5 m-Na+, to avoid chainlength dependent electrostatic effects) yields γ, the equilibrium constant for closing the minimum-size loop: γ = 0.003. Interpretation of the equilibrium constant for loop closure is discussed; it can be expressed as the equilibrium constant for formation of an isolated base pair in a bimolecular reaction multiplied by the effective concentration of one base in the vicinity of the other, before the loop is closed. The effective concentration is related to loop size by the loop-weighting function. The melting curves of closed hairpin helices can be used to test the formulation of the loop-weighting function. The results show that for small DNA loops the loop-weighting function differs from that predicted by Jacobson & Stockmayer (1950) for long gaussian chains; the differences are of the type expected for short chains with hindered rotation. Although the closed hairpin helices contain two loops and the open hairpins have only one, nevertheless the closed hairpins are more stable. The reason is a large difference in the conformational entropies of the two random-chain forms: the open circle can assume only a fraction of the conformations available to the linear chain. Since the midpoint of the melting curve (the Tm) depends on the relative stability of helix to random chain, the helix formed by the circular oligomer has the higher Tm. The Tm values of the closed hairpins are higher than predicted by earlier theories: higher, in fact, than the Tm of poly d(AT). The reason is that the minimum-size loops in a closed hairpin helix form more readily than use of the Jacobson-Stockmayer loop-weighting function would predict.