Endoscopic examination is a safe diagnostic method, however, it generates a large amount of data in the form of images that should be analyzed to detect unusual phenomena, such as bleeding. In this article, we propose a new adaptive algorithm for automatic detection of bleeding in endoscopic images. The proposed method begins with the image conversion to Hue-Saturation-Value matrix. The next steps involve the correlational fractal dimension calculation and adaptive bilevel thresholding of the hue channel. The threshold value is then adjusted according to the previous step as the blood smoothness (in comparison to the surroundings) is used in this calculation. This stage is followed by the image segmentation. Basing on this procedure, a simple decision on accepting or rejecting the blobs is taken. The decision-making process must also take into account the presence of areas that are similar in color to blood but do not represent bleeding. The entire procedure ends with a visualization of the blobs on full image.