The effective evaluation of Personal Information Management is a crucial problem for the research community. While evaluation methodologies for retrieval on the Web and in digital libraries are well-developed, the experiments with the advanced desktop tools are neither repeatable nor comparable. As privacy concerns do not allow to copy and distribute personal data outside the research lab, we suggest to overcome this problem by creation of desktop datasets within different research groups using a single methodology and a common set of tools. A dataset can include not only a static snapshot of the desktop documents, but also the logs of user activity on the desktop within several last months. We present the structure of the required dataset, a set of implemented tools and a sample dataset collected within the L3S Research Center.