We describe a novel scalable clustering framework for streamlines obtained from diffusion tractography. Clustering is an attractive means of segmenting a large set of streamlines into anatomically relevant bundles. For most existing methods, however, the large datasets produced in high resolution or multiple subject studies are problematical. To achieve good scalability, our method repeatedly divides the data into subsets, which are then partitioned using hierarchical clustering. A final partition is obtained by recombining the subsets. In addition, the recombination scheme provides a consistency measure for cluster assignment of individual streamlines, which is used to clean up the final result. The clusters have good anatomical plausibility and we show that three clusters corresponding to the three known segments of the arcuate fasciculus show excellent agreement with literature. A major advantage of the method is the fact that it can find clusters in datasets of essentially arbitrary size. This fact is exploited to find consistent clusters in concatenated tractography data from multiple subjects. We expect the identification of bundles across subjects to be an important application of the method.