When faced with an interface to an unknown system or device humans adopt exploratory interactive behaviour in order to gain information and insight. This paper describes a computer program which probes, observes and models the input–output space of unknown systems. We use a schema concept as the memory structure for recording events and adopt a constructive approach that avoids preprocessing the raw data. We believe qualitative assessments are important in early analysis and employ techniques from qualitative reasoning research in order to capture correlation behaviour. The aim is to gain insight into the nature of an unknown system for guidance in future model selection. A series of experiments and their results are discussed, together with the assumptions and limitations of the method. We suggest further developments for future experiments that appear promising.