Serum magnesium concentration has a neuroprotective effect in experimental models of traumatic brain injury (TBI). This study was designed to assess the relationship between initial serum magnesium, cerebrospinal fluid (CSF) magnesium, neurological outcome and the efficacy of magnesium replacement therapy (MgSO4). A retrospective analysis was performed on a prospectively collected dataset from 216 patients admitted during 1996-2006 to the University of Pittsburgh Medical Center with severe TBI. Admission serum and CSF magnesium were dichotomized into low and normal magnesium concentration groups for serum and normal and high concentration groups for CSF. A logistic-regression analysis was performed with 6-month Glasgow Outcome Scale (GOS) scores as outcome variable. The outcome of a subset of 31 patients who presented with low serum magnesium and who were rapidly corrected within 24 h of admission was also analyzed. Low initial serum magnesium was measured in 56.67% of all patients. Patients with an initial serum magnesium of <1.3 mEq/L were 2.37 times more likely to have a poor outcome (CI: 1.18-4.78, p = 0.016). The prognostic significance of depressed serum magnesium remained, even in patients whose serum magnesium levels were corrected within 24 h (OR = 11.03, CI: 1.87-68.14, p = 0.008). Patients with an initial high CSF magnesium were 7.63 more likely to have a poor outcome (p = 0.05). Elevated CSF magnesium correlated with depressed serum magnesium only in patients with poor outcome (p = 0.013). Patients with low serum magnesium and high CSF magnesium are most likely to have poor outcome after severe TBI. Rapid correction of serum magnesium levels does not reverse the prognostic value of these markers.