In order to make the detector perform robustly

an intuitive solution can be obtained by censoring the clutter observations from large and small deviations. In recent studies

censoring techniques have beenimplemented using order statistics （OS） and trimmed mean （TM） methods. In order to ultilize censored data

wepresent a quasi-best weighted （QBW） order statistics method resulting in an improved detection performancewhile censoring outliers. Based on quasi-best weighted method

a new greatest selection detector （QBWGO-CFAR） is proposed in this paper. Both Of its leading and lagging windows use QBW method to create twO localnoise power estimations

the greatest is selected as a globle estimation to set an adaptive threshold. Under the assumption of Swerling Ⅱ target and Rayleigh distributed clutter

the analytic expressions of Pfa

Pd

ADT and thepeak of false alarm at clutter edges are derived. The analytic results show that an improved performance overGOSGO or OSGO is obtained both in homogeneous background and in nonhomogeneous environment caused bystrong interfering targets and clutter edges. In specific cases. QBWGO-CFAR reduces tO GO and MX-CMLD.