Primitive human hematopoietic progenitors containing a high proportion of long-term culture-initiating cells (LTCICs) are found in the 34+DR- fraction of bone marrow mononuclear cells (BMMNCs). These progenitors adhere selectively to the 33/66-kD binding domain of fibronectin and to the FN-CHII binding site, unlike more committed progenitors, which adhere less selectively to fibronectin components. These differences in adhesion to stromal components may explain selective homing and release of progenitors at varying levels of differentiation from the marrow compartment. In additional studies, we demonstrate that cultivation of primitive progenitors in a stroma noncontact long-term culture allows both differentiation of primitive progenitors and conservation of LTCICs. These observations (1) demonstrate that expansion of primitive progenitors does not require stromal contact, (2) shed light on the regulatory role of stroma in myeloid differentiation, and (3) suggest strategies for both ex vivo myeloid progenitor expansion and retrovirus gene insertion. Finally, we demonstrate that a natural killer cell population can be derived from primitive hematopoietic progenitors in a modified long-term culture model. Our findings suggest an important role for marrow stroma in lymphoid differentiation from primitive progenitors and in expression of CD2, a lymphoid marker ordinarily associated with passage of T-lymphocyte progenitors through the thymus.