We studied nuclear morphometry of human breast cancer with special emphasis on two sources of variation: freezing prior to fixation and selection of measured objects on the basis of different sampling rules. Samples of 147 histologically verified invasive breast cancer cases were examined with a computer-based image overlay drawing system. Thirty-eight of the 147 samples of tissue frozen before embedding in paraffin were analyzed separately. Among the latter we found shrinkage of 35% and 46% (depending on the sampling rule) of the nuclear profile area as compared with samples not frozen before the standard tissue processing. These findings confirm that nuclear morphometry results from frozen and unfrozen tissue are not comparable. Frozen tissue later embedded in paraffin should not be used with prognostication models based on traditionally fixed tissue. In morphometry we applied two sampling rules that differed in the criteria used for selecting nuclei for measurement. We registered a significant difference in nuclear size and in the variation of nuclear size between the two sampling methods. Of the morphometric features studied, nuclear area was affected most. Finally, we examined the two sampling rules in light of the established prognosticators in breast cancer: tumor size, axillary lymph node status, and the Multivariate Prognostic Index (MPI). The two sampling rules resulted in different distributions of morphometric results in the prognostic groups. Our findings emphasize the significance of the sources of variation in nuclear morphometry. They also stress the need for well-standardized morphometric methods in predicting the outcome of breast cancer.