This article specifies the bases of an experimental design for comparing users’ performances in information retrieval (IR) with different systems. The field of the experiment concerns big collections of digital documents in a professional context like the companies’ Intranets. The classification schemes enable to show the knowledge shared in the company from different point of view. The aim of the experiment is to confirm that scheme-based navigation within a topic map of the field implies better performance in IR compared with a search engine. This article proposes a model of IR to activity and a first empirical study. Results enable to validate the experimental design and clarify the interpretation principles for further experiments.