In this paper

we present an improved constant false alarm rate（CFAR）algorithm for ship detection in syn-thetic aperture radar（SAR）imagery.The algorithm includes the probabilistic neural networks（PNN）

CFAR tech-nique

golden section method and area growth method.The PNN is used to estimate the probabilistic density function of radar backscatter from sea surface.The CFAR technique is applied to determine a threshold that differs ships form sea surface.The golden section method is used to estimate the shape parameter of the Gauss function while the area growth method is employed to remove the false alarm.The algorithm is applied to detect ships in Radarsat imagery.The compar-ison of the performance between the improved algorithm and the original algorithm is made.The results show that the im-proved CFAR algorithm performed better than the original one.