This paper describes a methodology that aims to detect and diagnosis faults in lightning arresters, using the thermovision technique. Thermovision is a non-destructive technique used in diverse services of maintenance, having the advantage not to demand the disconnection of the equipment under inspection. It uses a set of neuro-fuzzy networks to achieve the lightning arresters fault classification. The methodology also uses a digital image processing algorithm based on the Watershed Transform in order to get the segmentation of the lightning arresters. This procedure enables the automatic search of the maximum and minimum temperature on the lightning arresters. These variables are necessary to generate the diagnosis. By appling the methodology is possible to classify lightning arresters operative condition in: faulty, normal, light, suspicious and faulty. The computacional system generated by the proposed methodology train its neuro-fuzzy network by using a historical thermovision data. During the train phase, a heuristic is proposed in order to set the number of networks in the diagnosis system. This system was validated using a database provided by the Eletric Energy Research Center, with a hundreds of different faulty scenarios. The validation error of the set of neuro-fuzzy and the automatic digital thermovision imagem processing was about 10 percent. The diagnosis system described has been sucessefully used by Eletric Energy Research Center as an auxiliar tool for lightning arresters fault diagnosis.