Probabilistic Description Logics are the basis of ontologies in the Semantic Web. Knowledge representation and reasoning for these logics have been extensively explored in the last years; less attention has been paid to techniques that learn ontologies from data. In this paper we report on algorithms that learn probabilistic concepts and roles. We present an initial effort towards semi-automated learning using probabilistic methods. We combine ILP (Inductive Logic Programming) methods and a probabilistic classifier algorithm (search for candidate hypotheses is conducted by a Noisy-OR classifier). Preliminary results on a real world dataset are presented.