This paper presents a fast and robust algorithm for significantly reducing the number of false detections caused by screening algorithms for side scan sonar (SSS) images. The presented algorithm consists of two processing steps. First, an iterative segmentation process is carried out for seperating the image into shadow and background. This segmentation is based on an energy function which combines the local neighborhood segment information and the amplitude of a pixel. By minimizing this function, a clear shadow, the most significant target characteristic, will be extracted. Second, based on the region of interest (ROI) and the shadow contour, a robust classification approach is applied, utilizing the area of the shadow, the first two statistical moments of the pixel amplitude and the existence of parallel lines (Hough transformation). This algorithm was tested using a data set with approx. 2400 ROIs containing about 200 targets and 270 targetlike stones or sandrippel. This data set was gathered during five different measurement campaigns in the Baltic Sea and the Mediterranean Sea using three different SSS systems Benthos C3D, Klein2000 and Marine Sonics). These data were collected by FWG and WTD71 as well as by Atlas Electronik with the SeaOtter MK1 AUV.