Immunomagnetic sorting of natural killer (NK) cells from the peripheral blood of healthy donors has been evaluated in a comparative study of composition, yield and activation of target cells obtained by positive (Dynabeads, Microbeads) and negative (Microbeads) sorting procedures. Positively sorted target cells were selected by expression of the NK cell marker CD56, whereas NK cells obtained by negative sorting were those remaining after steps to remove all non-NK cell leukocyte populations were accomplished. In positive sorting, both CD56+CD3- NK cells and CD56+CD3+ natural killer T (NKT) cells were included. The NKT cell fraction differed between individuals, but not between the positive sorting methods. Whereas 20-30% of positively sorted target cells were NKT cells, only approximately 3% of negatively sorted cells were CD3+. Contamination with monocytes and B cells was low (1-3%) in all methods studied. Sorting with Microbeads (both positive and negative) gave higher cell yields than those obtained with Dynabeads (14% vs. 5% of total leukocyte numbers). A higher CD56 fluorescence intensity of NK cells and a better discrimination between the CD56bright and CD56dim NK cell subpopulations was obtained after negative sorting. Dynabeads-separated cells had, shortly after separation, a significantly higher expression (approximately 30%) of the early activation marker CD69 than cells either positively or negatively separated by Microbeads (approximately 8%). CD56+ cells positively sorted by Microbeads demonstrated a significantly higher production of TNF-alpha and IFN-gamma after IL-2 stimulation than Dynabeads-sorted cells. However, the cytotoxicity of cells obtained by the two positive sorting procedures did not differ. In conclusion, positive selection of CD56+ cells by Microbeads is better than Dynabeads, as determined from cell yield and procedure-associated cell activation and should be chosen for in vitro studies of NK/NKT cells. However, when pure NK cells and phenotypic subtypes are to be studied, negative sorting seems most appropriate.