The slow and uncertain convergence of multilayer feedforward neural networks using the backpropagation training algorithm is caused mainly by the iterative nature of the dynamic process of finding the weight matrices with static control parameters. This study investigates the use of fuzzy logic in controlling the learning processes of such neural networks. Each learning neuron in the neural networks suggested here has its own learning rate dynamically adjusted by a fuzzy logic controller during the course of training according to the output error of the neuron and a set of heuristic rules. Comparative tests showed that such fuzzy backpropagation algorithms stabilized the training processes of these neural networks and, therefore, produced 2 to 3 times more converged tests than the conventional backpropagation algorithms. The sensitivities of the training processes to the variations of fuzzy sets and membership functions are examined and discussed.