Most brain functions involve neuronal population activity that is distributed across multiple areas. The routing of signals through this distributed network is flexible, changing from moment-to-moment to meet task demands. To determine how flexible cortical communication could be instantiated, we recorded spiking activity of neuronal populations across several stages of the macaque cortical visual stream.  Using dimensionality reduction methods, we find that inter-areal interactions occur through a communication subspace: downstream fluctuations are related to a small subset of source population activity patterns. Subspaces for feedforward and feedback interactions appear distinct. We propose that the communication subspace may be a general, population-level mechanism by which activity can be selectively and flexibly routed across brain areas.