Which interaction technique works when?

We conducted an experiment that compared three post-WIMP interaction techniques: floating palettes, marking menus and toolglasses, in a real-world Coloured Petri-Net editor, CPN2000. We created six situations in which users performed identical sets of actions with equally-complex nets, but with different cognitive contexts. We found significant differences in performance and preferences across interaction techniques. When a user is in a "copy" context, floating palettes are more efficient. If the user is problem solving, toolglasses or marking menus are preferred. No single interaction technique is clearly superior: each has strengths in different contexts. Since a single application must support different kinds of cognitive tasks, interaction designers should consider integrating multiple interaction techniques, rather than selecting only one.