In this paper we discuss a model theoretic approach to learnability in predicate logic. Results in this direction were obtained by Osherson, Stob and Weinstein. It is assumed that there is a first-order structure given. Instances are tuples of elements of the universe of the model, and concepts are relations that are definable in the model by formulas from some pre-specified class. The goal of the learner is to identify an unknown target concept in some specific model of learning.