The widespread use of intensive therapies and the need to haematologically monitor patients on a frequent basis means that the proportion of blood samples with moderate to severe leucopenia is significant and increasing. From a laboratory perspective, particularly because of the need to spend significant amounts of time in obtaining manual differentials from stained smears with low leucocyte numbers, these clinical trends have created additional pressures on what is often a limited manpower resource. Moreover in such situations, differentials obtained from examination of only 20 or 50 cells are not uncommon and the statistical consequences of this will be clearly apparent. Currently, there is general user confidence for automated leucocyte differentials for blood samples with normal WBC parameters, but there has been some reluctance to extend this to samples with leucopenia. In order to explore this further, we examined the efficiency of a modern automated five-part differential analyser (Abbott CELL-DYN 3500) in an unselected series of 292 samples with leucopenia (WBC count range range; 0.28-2.48 x 10(9)/l). Of these, 49 were from leucopenic sero-positive HIV patients with the remaining 243 samples originating from haematological oncology clinics, patients receiving radiotherapy for non-haemopoietic malignancies, and from patients with various chronic diseases. Morphologically, 204 of these samples did not show any blast cells or NRBC, 48 had blast cells but no NRBC, 29 had NRBC but no blasts, and the remaining 11 showed both blasts and NRBC. For 277 cases with less than 5% blasts, there was an excellent correlation between the manual and CD3500 automated differential, with no obvious bias between manual and automated subpopulation estimates at any percentage level. Linear regression analyses comparing absolute neutrophil, eosinophil, lymphocyte and monocyte counts for these same samples further revealed impressive correlations (r > 0.92) for all leucocyte populations and the absolute neutrophil count in particular (r = 0.986). Manual and CD3500 leucocyte differential comparisons for 11 cases with > 5% blasts showed good correlations for absolute neutrophil and eosinophil counts although, when the blast cell percentage was high, correlations for lymphocyte and monocyte counts were less consistent (an operator alert in the form of a 'Blast Flag' was, however, given in 10/11 of these particular cases). Four additional cases where manual differentiation between lymphoid cells and monocytes was recorded as difficult also showed consistently good correlations for manual vs automated neutrophil and eosinophil estimates. Not surprisingly, and essentially as a result of the low confidence noted for the manual differential itself, correlations for lymphoid and monocytic cells were relatively poor. In conclusion, this study has demonstrated that the CD3500 provides reliable and accurate absolute neutrophil and eosinophil counts in leucopenic samples irrespective of the presence of blasts or NRBC. These observations are particularly important in terms of monitoring patients who are liable to develop neutropenia as a result of chemotherapy and radiotherapy, and provide evidence that the routine use of automated leucocyte differentials may be confidently extended to the analysis of leucopenic samples.