In this study, dielectric properties of polyesters which have abundant use in communication cables and power systems are aimed to be determined by means of adaptive neuro-fuzzy inference system (ANFIS). Measuring the dielectric properties of insulators under various fields of application and operating conditions requires advanced equipments and takes a long time. After measuring in a wide range and different conditions, determining the other required dielectric properties by means of ANFIS will supplement time and cost. To this end, ANFIS models are trained which are built for every measurement value of dielectric permittivity (e,) and loss factor (tans) with respect to heat-frequency values of polyesters used as insulators. Then, used data in ANFIS and unused data are tested separately. In the models built for determining dielectric permittivity and loss factor, absolute mean percent error-cumulative absolute percent error are found to be 0.0030-0.1092% and 0.0021-0.0766%, respectively. The results of ANFIS are compared with a previous study of multilayer perceptron (MLP) artificial neural networks (ANN) in which the same data sets are used. ANFIS models exhibit good learning precision and generalization according to mean absolute errors (MAEs). Hence, it becomes easy to determine the dielectric permittivity and loss factor without measuring the given temperature-frequency values by ANFIS. The percent effects of dielectric permittivity and loss factor that are determined by ANFIS on total dielectric loss are discussed based on representative analyses. (c) 2007 Elsevier B.V. All rights reserved.