This paper introduces Fuzzy ARTRON as a general-purpose classifier that can do high-quality classification in continuous, discrete, linear, or nonlinear domains. The topology of Fuzzy ARTRON contains a fuzzy ART network, on which a perceptron layer is superimposed. The learning algorithms involve unsupervised ART learning and supervised error back-propagation learning. The former is used to auto-construct proper clusters through the fuzzy ART self-construction ability. This improves the convergence rate and alleviates the local minima problem usually associated with the error back-propagation learning network. The latter is used to dynamically associate clusters with proper classes via connection weight adjustment. This improves the generalization ability so that Fuzzy ARTRON can successfully handle the linearly nonseparable problems usually associated with fuzzy ART and the weak generalization problem usually associated with fuzzy ARTMAP. Finally, Fuzzy ARTRON employs fuzzy hyperboxes to do clustering, which leads to better generalization performance compared to conventional hyperboxes. Computer simulations were conducted to evaluate the performance and applicability of Fuzzy ARTRON under continuous, discrete, linear, or nonlinear domains.