To be applied to practical clinical research problems, medical image computing software requires infrastructure including routines to read and write various file formats, manipulate 2D and 3D coordinate systems, and present a consistent user interface paradigm and visualization metaphor. At the same time, research software needs to be flexible to facilitate implementation of new ideas. 3D Slicer is a project that aims to provide a platform for a variety of applications through a community-development model. The resulting system has been used for research in both basic biomedical and clinically applied settings. 3D Slicer is built on a set of powerful and widely used software components (Tcl/Tk, VTK, ITK) to which is added an application layer that makes the system usable by non-programmer end-users. Using this approach, advanced applications including image guided surgery, robotics, brain mapping, and virtual colonoscopy have been implemented as 3D Slicer modules. In this paper we discuss some of the goals of the 3D Slicer project and how the architecture helps support those goals. We also point out some of the practical issues which arise from this approach.