Neural networks and decision trees are two common approaches to pattern recognition. In this paper, these approaches are combined to develop a new neural network architecture based on decision trees and a new learning rule to grow this architecture using neural network techniques. The resulting neural network is called a neural tree network (NTN). The NTN can be implemented very efficiently as compared to multilayer perceptrons (MLP). The learning algorithm is more efficient than the exhaustive search techniques used in standard decision tree methods. The algorithm also grows the network, thus finding the correct number of neurons as opposed to the backpropagation algorithm used to train MLPs in which the number of neurons and their interconnections must be known before learning can begin. Two different approaches are presented to grow the NTN based on self-organizing clustering techniques and a supervised learning rule. Simulation results are presented on a speaker-independent vowel recognition task which show the superiority of the NTN approach over both MLPs and decision trees.