ABSTRACT. A performance evaluation was conducted on a differential mobility classi er that separates bers according to length using dielectrophoresis. The classi er had been constructed and used for several applications in previous studies. The performance of the classi er was predicted using a two-dimensional axisymmetric model of the ow eld and then calculating particle trajectories for a variety of conditions. Based on the ow calculations, several regions of the classi er were improved to reduce likelihood of turbulent losses. For a given total ow through the classi er and a maximum voltage across the electrodes, the performance of the classi er was found to depend on the ratios of the aerosol ow to the inner and the outer sheath ows. It was found that the minimum classi able length, the minimum length distribution width, and the throughput of classi ed bers can each be optimized, but not independently. Several approaches to testing the resolution of the classi er were tried. The rst was to measure the length distribution of bers passing through the classi er under different conditions using electron microscopy. However, this was a slow and imprecise measure of performance. Two approaches using monodisperse latex spheres were used; one operated the instrument as an electrical mobility (electrophoresis) analyzer and the other evaluated only the ow system accuracy. All measures indicate that the classi er operates close to theoretical performance, but improvements are still possible. Suggested improvements require redesign of the ow system and improved electrode alignment.