Summary: The merit of automatic graph layout algorithms is typically judged by their computational efficiency and the extent to which they conform to aesthetic criteria (for example, minimising the number of crossings, maximising orthogonality). Experiments investigating the worth of such algorithms from the point of view of human usability can take different forms, depending on whether the graph has meaning in the real world, the nature of the usability measurement, and the effect being investigated (algorithms or aesthetics). Previous studies have investigated performance on abstract graphs with respect to both aesthetics and algorithms, finding support for reducing the number of crossings and bends, and increasing the display of symmetry.This paper reports on preference experiments assessing the effect of individual aesthetics in the application domain of UML diagrams. Subjects' preferences for one diagram over another were collected as quantitative data. Their stated reasons for their choice were collected as qualitative data. Analysis of this data enabled us to produce a priority listing of aesthetics for this domain. These UML preference results reveal a difference in aesthetic priority from those of previous domain-independent experiments. Communicated by Michael Kaufmann: submitted April 2001; revised December 2001 and June 2002.