Paleo-shorelines and ancient lake terraces east of Lake Manyara in Tanzania were identified from the backscatter intensity of TerraSAR-X StripMap images. Because of their linear alignment, edge detector algorithms were applied to delineate these morphological structures from those Synthetic Aperture Radar scenes. Due to the physical properties of microwave signals, this application has proven to be a challenging task for edge detectors. This study compares the performance of different combinations of speckle reduction techniques and edge operator in detecting linear paleo-shorelines. The Roberts, Sobel, Laplacian of Gaussian and the Canny edge detector algorithms were applied to extract and revise those linear structures. The comparison shows that the Canny edge detector is especially suitable for images with strong speckle noise. Canny achieves relatively high accuracies compared to the other operators. The stronger the filtering and speckle noise reduction, the better the performance of the other edge detection operators, compared to the Canny edge detector. The application of a wavelet transformation reduces the presence of artifacts resulting from speckle noise and emphasizes the detection of the target features.