Objective: The segmentation of biological metastatic volumes of tumor in PET is usually performed by threshold methods.This study was aimed to find an algorithm for calculating threshold in FDG-PET-based delineation of tumor volumes through phantom measurements and validated in patient data.Methods: PET scans were performed (Discovery LS 4) on a special phantom contains plastic column spheres with different diameters and filled with identical FDG concentrations.Gradually change the activity of the FDG solution with different signal-to-background ratios form 2∶0, 2∶1 to 20∶1.By analyzing the images we got form phantoms, we got four different variables, the SUV max of each spheres (SUVmax), the SUV in the border of each spheres(SUVborder), the mean SUV of a 1cm×1cm interest region in background(SUVbg), the diameter of spheres(D).Using SPSS 13.0 to find the lineation relationship between those variables, we founded an algorithm to calculation the threshold for FDG-PET-based delineation of tumor volumes, and validated the algorithm in 35 lung cancer lesions (21 primary lung cancer, 14 metastatic lymph nodes) with pathology confirmed.The results of PET volumes were calculated based on the algorithm, and than compared with CT volumes of tumor which is delineated in lung window.Results: Phantoms data analysis show us a algorithm: SUVborder =0.331 * SUVmax + 0.468 *SUVbg + 0.139 * SUVmax/D (cm), r =0.994.In 35 lesions validation, the average gross tumor volumes (GTV) delineated by PET and CT are 7.36±1.62ml and 8.31±2.05ml, and no statistically significant difference between them.Conclusion: By the algorithm we presented in this study, a FDG-PET based delineation of Tumor is viable.The threshold is related to both the SUVmax and size of tumor, and also the SUV of background.By comparing to CTs GTV, PETs GTV is smaller, but no statistically significant difference was found.