We present a method of learning relationships at the triadic level of a relationship network. The method proposes learning linkages of a particular network using a support vector machine (SVM) classifier trained on the known part of a relationship network. Using features drawn from the topological information of the two degrees of separation of a link a classifier learns whether two people of that link are related or not. We investigate empirically the performance of the technique for various relationship networks derived from email, web hyperlinks, and questionnaires