A new, practical, and informative technique is presented for the efficient calculation of classifier efficiencies. Kernel density estimation, a well-developed statistical tool, is applied to particle-size datasets from classifier flow streams. It is demonstrated that this construction of particle-size densities is more accurate and appropriate than histograms. The resulting classifier efficiency curves then display the key features of the classifier efficiency without propagating the noise inherent in most particle-sizing techniques, and without the constraint of any selectivity curve parametrization.