The reliability of fine-needle aspiration cytology (FNA) for distinguishing between carcinoma, lymphoma, and sarcoma was established in a previous study (Thunnissen et al., Cytopathology 1993; 4:107-114). The purpose of this study was to investigate which criteria were useful for a probabilistic diagnosis. A total of 78 randomly chosen FNA smears (31 carcinomas, 24 lymphomas, and 23 sarcomas) was sent around and read "blindly" by six cytopathologists. Each pathologist completed a list of 16 criteria for every case. Histology was used as a reference standard. A statistical analysis led to the selection of three criteria: "lymphoglandular bodies," "well-defined clusters," and "spindle-cell nuclei," associated with lymphoma, carcinoma, and soft-tissue sarcoma, respectively. Given the scores on these criteria, the probabilities to be assigned to the three diagnostic categories can be read from a table. It turns out, as one might expect, that the classification of the most probable disease is pretty reliable if one cytologic criterion scores much higher than the other two criteria. On other cases, fuzziness appears and misclassifications are far from improbable. This study offers a general cytologic approach. The cytologic criteria "lymphoglandular bodies," "well-defined clusters," and "spindle-cell nuclei" can be used both in daily practice and in education to assign posterior probabilities to carcinoma, lymphoma, and soft-tissue sarcoma.