To perform an optimal ex vivo bone marrow purge, it is necessary to concentrate the bone marrow progenitor cells and to eliminate both the red blood cells and the polymorphonuclear leucocytes. To achieve this goal which cannot be accomplished by using the Haemonetics model 30 alone, we used the Haemonetics model 30 and a density gradient together in a two-step procedure. In the first step we obtained the buffy coat from original bone marrow grafts and in the second we reintroduced these buffy coats into the Haemonetics bowl followed by Percoll, adjusted to 1.079 g/ml, at 5 ml/min in order to recover the light density mononuclear cells. After this second step, the mean volume of the marrow and the RBC and nucleated cell contaminations were reduced to 9%, 0.96% and 16% of their original values the unseparated bone marrow, respectively. This was far better than the values obtained after the first step (volume: 19%; RBC: 10%; nucleated cells: 54%). The CFU-GM recoveries after the first and second steps were 71% and 70% of the original samples, respectively. The entire procedure lasted between 75 and 150 min. At this time, 17 of the 24 patients whose bone marrow was separated using Percoll gradient in the Haemonetics bowl have been grafted. Thirteen of these 17 patients had an evaluable haematological recovery which was complete and rapid for all but one patient with acute myeloid leukaemia. These results demonstrate that the introduction of a density gradient into the Haemonetics model 30 bowl is possible and effective. The reduction in total volume and cell number permits ex vivo purging, without decreasing the grafting capability.