In clinical cytology, nuclear features play an important role in cell and tissue classification. To increase efficiency and decrease subjectivity of cytological results, automation of the analytic process has been proposed and discussed by many authors. This automation can be achieved by estimating the probability of occurrence of a certain class given particular features of a microscope specimen. In this paper, feature sets that might be used as inputs for mathematical cytological classification algorithms are reviewer. The primary goal was to determine the important properties of these features sets, i.e., are there mathematically efficient features that provide a more or less compete description of the cell. Under what conditions will these feature then result in optimal classification of the cells using quantitative fluorescence staining. And how would these mathematical features relate to conventional features that a human observer understands. Example human observer features are size, shape, and chromaticity o the cell nucleus while example mathematical features are image moments. If the cell image can be completely reconstructed from the feature set, then it should be possible to derive the conventional features used by human observers from the mathematical feature set for presentation to clinicians. Finally, the suitability of different mathematical decision making algorithms like probabilistic reasoning, clustering or neural networks are also briefly evaluated in the context of a mathematically complete feature set.