In many problems in electromagnetics engineering, there is the possibility to know examples, namely input-output relationships, which enable the use of suitable and effective learning-by-examples (LBE) approaches in order to determine the unknown solution/output for a new set of input data. Such examples can be obtained from past experience or be synthetically defined. In this framework, approaches based on the use of Support Vector Machines (SVMs) have been widely applied in the last years also to many problems in electromagnetics. In this paper, a set of such SVM-based approaches is presented. Starting from an accurate description of the state-of-art, potentialities and limitations of SVM are pointed out and future trends are envisaged together with possible new applications.