A method is developed for the analysis of centrifugal filters that are applied to the removal of small particles from a carrier fluid. The particular application considered is the removal of entrained microscopic oil droplets from "blow-by" escaping from the crankcase of an internal combustion engine. The analysis is general for the type of filter geometry considered, and is applied to two particular vane shapes, one of which is circular, and the other of which maintains a constant angle between the vane blade and the outgoing radius; the analysis results in expressions for transverse distances travelled by the particles between the vanes within their residence time within the filter. It is shown that an optimum radius of curvature or blade angle exists in each case. These expressions are used to predict filter efficiency in 3 models: a direct deterministic model, a model that incorporates the statistical nature of particle movement, and a model that includes a measure of particles susceptibility to be maintained in suspension due to turbulent fluctuations within the flow. Experimental results are given for a range of particle sizes and filter rotation speeds. It is shown that the deterministic model estimates the speed at around 95% efficiency well, but that the statistically based model gives better prediction over a broad range of particle sizes and spin speeds. The inclusion of a term to measure the particles susceptibility to turbulence is shown to give much better prediction for sub-micron particle sizes. (C) 2010 Elsevier Ltd. All rights reserved.