The paper deals with detection of hot spots in thermograms acquired after infrared inspections of power transmission lines. The author proposes an approach to automated selection of a threshold value for image segmentation using several methods of image processing and analysis. The algorithm is based on finding the steepest growth of successive gradients of sorted values of input image after certain image pre-processing steps. Several post-processing procedures were applied for proper visualization of detected regions in the resulting images. The results have brought satisfying effects and the algorithm, owing to its fast performance, could be also used on-line, during vision inspections.