Data on the Swedish population are stored in many registers located in different authorities. These data are crucial for epidemiological and other register-based research. We are developing an infrastructure to support research on the data from Swedish registers. In the infrastructure each organization or authority should keep its own control on all accesses to its data. The infrastructure should allow only authorized access to data. The effort for scientists to perform their analyses on the data should be insignificant. We propose to base the infrastructure on federated databases. This enables secure and only authorized data access, efficient and scalable data extractions, integration with statistical and other tools, and autonomy to data sources. We are investigating the feasibility of our approach by developing a pilot infrastructure for epidemiological research on cervical cancer. In this paper we identify challenges of applying federated databases for accessing sensitive and personal data by scientists.