Many previous studies have subdivided auditory neurons into a number of physiological classes according to various criteria applied to their binaural response properties. However, it is often unclear whether such classifications represent discrete classes of neurons, or whether they merely reflect a potentially convenient but ultimately arbitrary partitioning of a continuous underlying distribution of response properties. In this study we recorded the binaural response properties of 310 units in the auditory cortex of anesthetized ferrets, using an extensive range of interaural level differences (ILDs) and average binaural levels (ABLs). Most recordings were from primary auditory fields on the middle ectosylvian gyrus and from neurons with characteristic frequencies >5 kHz. We used simple multivariate statistics to quantify a fundamental coding feature: the shapes of the binaural response functions. The shapes of all 310 binaural response surfaces were represented as points in a 5-dimensional principal component space. This space captured the underlying shape of all the binaural response surfaces. The distribution of binaural level functions was not homogenous as some shapes were more common than others. Despite this, clustering validation techniques revealed no evidence for the existence of discrete, or partially overlapping, clusters that could serve as a basis for an objective classification of binaural level functions. We also examined the gradients of the response functions for the population of units; these gradients were greatest near the midline, which is consistent with free-field data showing that cortical neurons are most sensitive to changes in stimulus location in this region of space.