Although the hydroxyl radical is known experimentally to strongly bind an electron, Hartree–Fock calculations predict the excess electron to be unbound. The electron affinity of hydroxyl radical is therefore entirely due to differential effects of electron correlation between the neutral and the anion. Provided that sufficient electron correlation is included in the wave functions, it is found that basis set requirements for semiquantitative determination of this property are modest. A standard double zeta Gaussian basis augmented by one shell of diffuse functions on each atom is capable of giving over 75% of the experimental electron affinity. Addition of polarization functions makes a small correction leading to recovery of over 80% of experiment. Convergence with respect to enlargement of the basis set is very slow beyond this point. Through comparison of a series of calculations containing different levels of configuration interaction, it is found that the electron affinity is largely due to certain types of double excitations from the dominant RHF-like reference determinant. One kind involves only intrapair double excitations and is properly regarded as representing intrapair correlations. The other kind, which is just as important, involves products of two single excitations from different pairs. These then represent interpair correlations arising from simultaneous intrapair single excitations. The GVB method leads to poor results, due to the neglect of the latter kind of correlation. The pi electrons give most of the EA, the interaction of the pi electrons with the sigma bonding pair makes a small but significant contribution, and the 1s and 2s oxygen pairs have little effect. Based on this, a simple MCSCF model including only intrapair excitations is found that leads to very good results for the electron affinity without the necessity of obtaining a large share of the total correlation energies. Further refinements to include higher order intra- and interpair effects via complete active space MCSCF have little effect and even large scale CI corrections are small.