Many applications require comparison between alternative scenarios; most support it poorly. A subjunctive interface supports comparison through its facilities for parallel setup, viewing and control of scenarios. To evaluate the usability and benefits of these facilities, we ran experiments in which subjects used both a simple and a subjunctive interface to make comparisons in a census data set. In the first experiment, subjects reported higher satisfaction and lower workload with the subjunctive interface, and relied less on interim marks on paper. Subjects also used fewer interface actions. However, we found no reduction in task completion time, mainly because some subjects encountered problems in using the facilities for setting up and controlling scenarios. Based on a detailed analysis of subjects' actions we redesigned the subjunctive interface to alleviate frequent problems, such as accidentally adjusting only one scenario when the intention was to adjust them all. At the end of a second, five-session experiment, users of this redesigned interface completed tasks 27% more quickly than with the simple interface.