This paper is concerned with data modelling and the question of how to improve the quality of software tools to provide more effective support for the modeller. With the increasing interest in building intelligent interfaces to databases and associated support tools, we thought it instructive to embrace human-computer interaction (HCI) concepts, particularly ideas of user needs analysis and methods of interface evaluation. In the pilot study reported here, we step back from implementation issues to learn more about what goes on during the modelling process. Three competent student modellers were filmed giving a spoken commentary as they tackled an on-line modelling task. Two subjects, ST1 and ST2 used a software tool developed in-house, while subject FC modelled the same scenario using a flip-chart. Analysis of the resulting video film yielded a number of interesting insights into how the three subjects approached the task, and in particular, about how their medium helped or hindered. In this report we concentrate mostly on subjects ST1 and FC who proved to be well matched in ability, neither evidencing any conceptual difficulties. The observations made from this preliminary experiment confirm the untapped potential of this technique among the database community. Moreover, on the basis of the findings, it seems that we have much to learn from going back to pen and paper, about the features a successful data modelling tool should provide.