Separation is a bottleneck problem in many colloidal fabrication operations. Particles in the size range 10 nm to 10 micrometers are notoriously difficult to separate from each other by size or shape. Three important considerations are 1) the required rate of separation for a given cost, 2) the fragility of the “particles” (e.g., silica spheres, DNA, colloidal assemblies), and 3) the required separation accuracy. In our lab the critical separation requirement is in separating properly-formed colloidal assemblies from unreacted pre-cursors or junk. A seemingly obvious and scalable way to separate suspended particles by size or shape is centrifugation or density gradient centrifugation. Indeed the technique has advantages in robustness, simplicity, and cost over other techniques for most separations. However, a notorious instability problem occurs during density gradient centrifugation, such that the particles simply mix throughout the centrifugation vessel. In this talk we show that the instability is a Rayleigh-Bénard instability. The theory then suggests to us a new technique that provides improved separation ability for colloidal particles.