The analysis of serial crime is based on a broad variety of data coming from different sources; for instance, the analysis can be performed through the use of spatio-temporal reasoning processes, inferences that use forensic science data, as well as the evaluation of modus operandi. The computerised extraction of knowledge from databases (data mining) is often seen as a promising approach to treat the available data. It is argued that this will only succeed if a modelling process is initialised, primarily aimed at a better understanding of the real nature of crime analysis and of the complex environment in which such systems have to be integrated. Focusing on knowledge of investigators is proposed as a first useful approach in that perspective.