We present rules that allow one to predict the stability of DNA pyrimidine-purine-pyrimidine (Y-R-Y) triple helices on the basis of the sequence. The rules were derived from van Prediction of macromolecular physical properties using only the sequence is one of the primary goals of biophysical chemistry. The utility of such models is in the ability to predict the stability of RNA or DNA helices based on the sequence alone (1, 2), the flexibility of the DNA duplex (3), local helical geometry (4,5), and the propensity of a particular sequence to undergo a structural transition between different helical forms (6). A predictive model for triple-helix stability could have broad application as well. Triple-helix formation can be used to recognize DNA duplexes highly specifically (for reviews, see refs. 7-10) and has potential for antisense and therapeutic applications. In this paper, we present a model for predicting DNA triplex stability using only the sequence.In developing a predictive model, two features are important: (i) the appropriateness of the parameters used to construct the model and (ii) the distribution of the sequences used as the basis set in its parameterization. For the prediction of the enthalpy of triplex formation (AH°), a nearest neighbor model was used, whereas a combination model (one containing a mixture of mono-and dinucleotide parameters) was found to be best for prediction of the free energy of the triplex (AG° [Na+] cacodylate/cacodylic acid, 1 mM EDTA (experiments at pH 6.25-7.0). The Na+ concentration was held constant at 100 mM because it is known to affect the triplex tm (10, 12) Melting experiments were performed by heating from low temperature to high temperature at -0.5°C per min and data were taken every 0.5°C. tm values were reproducible +0.5°C.Determination of Triplex Stability. The stability of the 23 intermolecular triplexes was determined by van't Hoff analysis of absorbance melting curves over a pH range from 4.75 to 7.0. Between pH 4.75 and 5.5, curves were analyzed as described (11,13). Briefly, the melting data were treated with a statistical mechanical approach where the equilibrium is broken up into discrete states (11,14)