To determine whether the histologic lesions classified by the system of Arlet et al as Type 2 (granular necrosis of fatty marrow) and Type 3 (complete medullary and trabecular necrosis) always progress to Type 4 (complete necrosis with marginal medullary fibrosis and appositional new bone formation), 10 femoral heads (nine patients) were monitored for 4 years using serial magnetic resonance images. These femoral heads had been diagnosed histologically as having either Type 2 (seven hips) or Type 3 (three hips) necrosis on initial core biopsies. On the initial magnetic resonance image, none of the femoral heads showed any focal lesions indicative of osteonecrosis. In all instances, superselective angiography showed interruption of the superior retinacular artery, and the bone marrow pressure was elevated. During a followup period of 48 to 54 months, no patient had a reactive low signal intensity band develop on T1 weightings, as evidence of a reparative process around the necrotic portion of the lesion, or any other findings of osteonecrosis on magnetic resonance images. These findings suggest that some Type 2 and 3 lesions of Arlet et al may not develop an obvious reactive interface of reparative revascularization and thus may not progress to definite and classic Type 4 osteonecrosis. This study supports the hypothesis that there is an ischemic threshold between reversible intraosseous hypoxia (bone marrow edema syndrome) and irreversible intraosseous anoxia (classic bone infarction or osteonecrosis) and suggests that borderline necrosis occurs in the transition zone of this ischemic threshold and is nonprogressive.