Background: After traumatic brain injury (TBI), plasma concentration of glial fibrillary acidic protein (GFAP) correlates with intracranial injury visible on CT scan. Some patients with suspected TBI with normal CT findings show pathology on MRI. We assessed the discriminative ability of GFAP to identify MRI abnormalities in patients with normal CT findings. Methods: TRACK-TBI is a prospective cohort study that enrolled patients with TBI who had a clinically indicated head CT scan within 24 h of injury at 18 level 1 trauma centres in the USA. For this analysis, we included patients with normal CT findings (Glasgow Coma Scale score 13–15) who consented to venepuncture within 24 h post injury and who had an MRI scan 7–18 days post injury. We compared MRI findings in these patients with those of orthopaedic trauma controls and healthy controls recruited from the study sites. Plasma GFAP concentrations (pg/mL) were measured using a prototype assay on a point-of-care platform. We used receiver operating characteristic (ROC) analysis to evaluate the discriminative ability of GFAP for positive MRI scans in patients with negative CT scans over 24 h (time between injury and venepuncture). The primary outcome was the area under the ROC curve (AUC) for GFAP in patients with CT-negative and MRI-positive findings versus patients with CT-negative and MRI-negative findings within 24 h of injury. The Dunn Kruskal–Wallis test was used to compare GFAP concentrations between MRI lesion types with Benjamini–Hochberg correction for multiple comparisons. This study is registered with ClinicalTrials.gov, number NCT02119182. Findings: Between Feb 26, 2014, and June 15, 2018, we recruited 450 patients with normal head CT scans (of whom 330 had negative MRI scans and 120 had positive MRI scans), 122 orthopaedic trauma controls, and 209 healthy controls. AUC for GFAP in patients with CT-negative and MRI-positive findings versus patients with CT-negative and MRI-negative findings was 0·777 (95% CI 0·726–0·829) over 24 h. Median plasma GFAP concentration was highest in patients with CT-negative and MRI-positive findings (414·4 pg/mL, 25–75th percentile 139·3–813·4), followed by patients with CT-negative and MRI-negative findings (74·0 pg/mL, 17·5–214·4), orthopaedic trauma controls (13·1 pg/mL, 6·9–20·0), and healthy controls (8·0 pg/mL, 3·0–14·0; all comparisons between patients with CT-negative MRI-positive findings and other groups p<0·0001). Interpretation: Analysis of blood GFAP concentrations using prototype assays on a point-of-care platform within 24 h of injury might improve detection of TBI and identify patients who might need subsequent MRI and follow-up. Funding: National Institute of Neurological Disorders and Stroke and US Department of Defense.