Active learning is a key element of constructivist learning theory and has been used as an argument for employing discovery-based designs with instructional software. On the other hand, researchers have highlighted empirical evidence showing that 'pure' discovery-based learning is of limited value. This suggests that how learners interact is important in predicting whether learning occurs. This paper reports on a study of 158 university students who each used two instructional simulations - one with a discovery-based design and the other with a tutorial-based design. Students' learning outcomes were assessed via pre-tests and post-tests of conceptual understanding. Students' interactions using the discovery-based program were recorded and coded as either systematic or unsystematic. The results showed that when compared with the tutorial-based learning program systematic exploration resulted in learning benefits, while unsystematic exploration did not. These results have implications for the design of instructional software if such resources are to be used effectively.