Much of the information recorded at the time of surgery for malignant disease has been used, alone or in combination, to indicate the outlook for that patient, including probability of metastasis. However, direct evidence of tumour seeding in distant organs at that time is not available. An immunocytochemical assay for the epithelial cytokeratin protein (CK18) may fill this gap since it is a feature of epithelial cells but would not normally be in bone marrow. We looked for disseminated tumour cells preoperatively in bone marrow from 88 patients with radically resected colorectal carcinomas. Smears for 28 (32%) patients were positive for cells of epithelial origin, which were absent from aspirates taken from 102 controls with non-malignant disease. The prognostic value was assessed in a follow-up study with a median observation time of 35 (12-58) months. Patients who had bone marrow tumour cells in the aspirate showed a significantly shorter disease-free survival than those without tumour cells (p = 0.0084). In a Cox regression model multivariate analysis demonstrated that the finding of tumour cells in bone marrow is an independent, significant (p = 0.0035) determinant of relapse. Although the skeleton is not a preferred site of overt metastasis in colorectal cancer the demonstration of tumour cells in bone marrow has to be taken as evidence of the general disseminative capability of an individual tumour.