Detection of isolated tumour cells (TCs) in bone marrow (BM) from epithelial cancer patients by immunocytochemical (ICC) analysis seems to predict future relapse, but the reported percentages of positive BMs among patients with localized cancer show large variations and the number of detected TCs is low. This emphasizes the importance of thoroughly testing the methods in use. This study was performed to clarify to what extent positive staining of haematopoietic cells (HCs) interferes with the ICC detection of epithelial cells in BM. BM mononuclear cells (MNCs) from normal donors and stage I-II breast cancer patients were stained with anti-cytokeratin (CK) and isotype control monoclonal antibodies (MAbs) followed by alkaline phosphatase (AP)-based visualization of immunolabelled cells. In the ICC staining of normal donors by the anti-CK MAbs AE1/AE3 or A45-B/B3, rare immunoreactive cells were detected in 7/20 and 8/19 BMs, respectively. Morphological examination recognized all these cells as typical HCs. In the breast cancer patients (n = 257), anti-CK-positive cells were detected in 26.6 per cent, excluding cells with HC morphology. Using the same morphological criteria, isotype control-positive cells were detected in 5.4 per cent of patients. Some of the false-positive events were further analysed and cells with strong reactivity against the AP enzyme alone were detected. Double ICC staining recognized the majority of these AP directly-reactive cells as CD45-negative and human Ig kappa/lambda-positive, in accordance with the phenotype of mature plasma cells. Morphological evaluation and adequate controls are important to ensure the diagnostic specificity of micrometastases in BM. It is recommended that the number of BM MNCs included in negative controls should equal the number of cells in the diagnostic specimens.