Fluoromisonidazole (FMISO) has been shown to bind selectively to hypoxic cells in vitro and in vivo at radiobiologically significant oxygen levels. When labeled with the positron emitter fluorine-18 (F-18), its uptake in tissue can be detected quantitatively with high precision by positron emission transaxial tomography (PETT). This paper presents the first experiences with PETT imaging of [F-18]FMISO uptake in human malignancies, and describes the development of this technique as a tool for the non-invasive assessment of tumor hypoxia. Eight patients with selected cancers were imaged prior to primary radiotherapy, and 3 returned for follow-up scans, for a total of 11 imaging studies. Six of eight pre-radiotherapy studies revealed retention of [F-18]FMISO in tumors that significantly exceeded plasma concentrations by 2 hr after drug injection; all five patients with head and neck primaries had such "positive" scans. An analytic method for the interpretation of [F-18] FMISO PETT images is presented, defining hypoxic elements within a tumor volume as regions with a threshold regional tumor:plasma [F-18]FMISO ratio of greater than or equal to 1.4 by 2 or more hours after injection. Toward the end of a course of fractionated radiotherapy, three repeat studies in patients with initially positive scans showed no tumor accumulation of drug above the threshold ratio of 1.4, suggesting reoxygenation had occurred. Pharmacokinetic and dosimetry data support continued use of [F-18]FMISO as a safe hypoxia probe. Two imaging protocols have been developed for human studies; a long protocol allows for more complete biodistribution and dosimetry information, and a shorter protocol facilitates increased patient accrual by applying a simple, clinically expedient imaging procedure. When correlated with tumor outcome, [F-18]FMISO PETT imaging may be developed as a predictor of tumor response to conventional radiotherapy. The implications of this technique in addressing persistent questions of tumor hypoxia in human oncology is discussed.