To study the ability of dual phase FDG-PET/CT imaging to accurately distinguish tumor versus necrosis in patients treated for brain metastases. 32 (22 female, 10 male) consecutive patients with treated brain metastases, lesion size greater than 0.5 cm(3) and suspected recurrence on MRI underwent dual-phase FDG-PET/CT. Clinical outcome was assessed by biopsy or by MRI. SUVmax and SUVmean values of the lesion (L) and gray matter (GM) at the level of the thalamus were measured on early (1) and delayed (2) imaging. L1/GM1 and L2/GM2 and the change of L/GM ratios as a function of time were calculated [(L2/GM2 - L1/GM1)/(L1/GM1)]. Cut-off values were obtained by ROC analysis. P < 0.05 defined statistical significance. Seven patients were excluded due to indeterminate outcomes. 25 patients (16 female, 9 male; 27 lesions; 28 scan sessions) had clear outcomes, proven by either biopsy (n = 16 patients) or serial follow-up MRI (n = 9 patients). Primary subtypes included breast (n = 9), lung (n = 7), melanoma (n = 3), squamous cell cancer of the head and neck (n = 2) and other (n = 4). Twenty-two patients underwent prior radiation (2-113 months) and three received only prior chemotherapy (5 months to 3 years). A change >0.19 of L/GM ratios as a function of time was 95% sensitive, 100% specific, and 96.4% accurate (P = 0.0001; AUC = 0.97) for distinguishing tumor versus radiation necrosis. The ratio of the change of the lesion to WM ratios over time was the second best indicator of outcome when compared to all indices used (ROC cut-off = 0.25, sensitivity 89.5% and specificity 90.9%, and accuracy 89.2%; P = 0.0001; AUC = 0.95), Early or late SUVs of the lesion alone did not differentiate between tumor and necrosis. Regardless of histological type, differentiation of necrosis from metastatic brain lesions was improved by using the change of lesion to gray matter SUVmax ratios as a function of time.