Fog computing is becoming a popular paradigm for bringing the advantages of the cloud nearer to the network edge. This way, computational tasks can be offloaded from end devices to nearby fog nodes, thus benefiting from high computational power and low latency at the same time. Architecture plays a central role in fog computing. Many papers on fog computing address architectural questions. However, a closer look reveals that different papers use the term “architecture” for very different concepts. This is rooted in the multi-disciplinary nature of the fog computing paradigm. The different communities involved in fog computing—network, hardware, system software, application software—all use the term “architecture,” but with different meaning. To facilitate the mutual understanding of architectural issues in fog computing, this paper introduces a conceptual framework for reasoning about architecture in fog computing. This conceptual framework uses three independent dimensions to describe architecture. Based on the three architecture dimensions, several architecture views can be defined to serve the different viewpoints of the involved disciplines, and to highlight different aspects of the architecture. The conceptual framework is validated using a literature mapping study.