This study compares some of the behavioural characteristics of two recommender systems for scholarly articles in a digital library: a usage-based recommender and an experimental citation-based recommender. Experimental results show that article recommendations based only on usage data are slightly better at solving the perennial data-sparsity problem that plagues collaborative filtering recommenders in digital libraries. However, citation-based recommendations are more semantically diverse and have less in common with conventional search results than the usage-based method. However both of these methods are complementary since most of the time if one recommender produces a list of recommendations the other does not.