Acoustic imaging of elastic targets often results in linear flashes in the image. Delineating these target features could be used in subsequent target detection and recognition schemes; however, the difficulty in this task is not to be underestimated due to the presence of background reverberation or ambient noise in the sonar image. A powerful line-fitting algorithm which locates the linear flashes in sonar images of elastic objects is presented. The algorithm initially groups the image pixels into neighboring clusters by using the parameters of an intensity-weighted least-mean-squared line fit locally computed over the entire image. The algorithm uses the information of these groups in order to extract the dominant lines from the image. It is noteworthy to realize that compared to the majority of line and edge detection schemes, the algorithm actually parametrizes for each line extracted its length, orientation, location, and intensity. By culling these lines based on their parameters, particular linear target features may be emphasized. Applying the algorithm to images produced from experimental high-frequency sonar data has shown that it performs even in consideration of acoustic noise.