Although cytological diagnosis plays a significant role in the management of thyroid cold nodules, the rather high rates of false negative cases diminishes its usefulness. The purpose of this preliminary study is to evaluate the utility of numerous morphological criteria used by the cytologist to exclude benign tumours. Thirty-one cytological criteria were routinely scored as binary (yes/no) or as categories: 6 referred to the general arrangement and frequency of thyroid cells, 9 to the associated cellular and cell product elements, and 16 to the morphological features of the cells. We examined the manner in which these criteria, alone or combined, contributed to the diagnosis. The data base consisted of 171 intraoperative imprint cytological samples (143 histologically benign, 1 atypical adenoma and 27 cancers), as well as 257 thyroid cold nodule aspirates from another set of patients (198 histologically benign, 7 atypical adenomas and 52 cancers). For the imprint cytology, the diagnostic power of each criterion was individually assessed by the likelihood ratio (LR) which eliminated 11 as being undiscriminatory. The remaining independent criteria were subjected to logistic regression analysis to determine the most discriminant. Three were selected: Cellular clustering organisation, nuclear hypertrophy and colloid quantity with the latter being somewhat less powerful. Furthermore, it appears that the diagnostic power of the criteria was significantly lower when there was at least one nucleolus (number of nucleoli greater than 0). The smears gave essentially the same results as the imprints.(ABSTRACT TRUNCATED AT 250 WORDS)