When people are presented with noun-noun compounds, they tend to produce two main types of interpretation: relational and property interpretations. One theory of compounding maintains that relational interpretations are preferred over property ones. However, many of the studies supporting this relational preference hypothesis appear to be vitiated by a failure to control for the familiarity of compounds. A rating study on so-called "novel" compounds used in previous studies is reported, which shows that many can be considered to be familiar. Then two experiments in which familiarity is controlled are presented, to test the relational preference hypothesis, using a sensibility judgment task (Experiment 1) and a comprehension judgment task (Experiment 2). The results show that familiarity has a clear effect on the ease of understanding of noun-noun compounds but that there is no hard evidence for relational preference. The implications of these results for the empirical literature and for current theories are discussed.