In the last decade, numerous studies in inductive reasoning have shown that people are remarkably quick at picking up the relevant features that allow generalizations. One formal account of (some instances of) cases where this ability emerges, relies on the assumptions people make about the manner in which the observations are sampled from the world. According to a weak sampling scheme, observations are randomly drawn from the environment. However, a reasoner can also assume that the observations are sampled deliberately from the intended hypothesis (strong sampling). In the present contribution, we compare these assumptions in a inductive reasoning task using semantic stimuli from 4 superordinate concepts (animals, clothes, vehicles,musical instruments). Model analyses are performed to examine the sampling assumptions of reasoners in this context. We find that people generally assume a strong sampling scheme. However, the model seems to underestimate the relevance sensitivity of the participants.