A version of the Generalized Context Model (GCM) (Nosofsky, 1984, 1986) with a deterministic response rule, and a decision bound model known as the general quadratic classifier (GQC) (Ashby & Maddox, 1993; Maddox & Ashby, 1993), were compared on their ability to predict the asymptotic trial-by-trial classification data in a series of experiments involving large, ill-defined category structures. In preliminary comparisons, it was shown that when the optimal classification boundary that separated the two experimental categories was quadratic, the models provided roughly equivalent accounts of the data. In a pair of experiments in which the optimal classification boundary was of a more complex form than quadratic, the deterministic GCM significantly outperformed the GQC. Several decision bound models that postulated more complex decision boundaries than the GQC were developed and tested, including a connectionist learning model that implements quartic polynomial decision boundaries. The deterministic GCM received strong support in Experiment 1, outperforming all of the decision boundary models tested. Experiment 2, however, suggested possible limitations in human classification abilities not predicted by the GCM. Implications of these findings for decision bound and exemplar models were discussed.