Abstract: Disclosed is an improvement in methods for reading bar codes, enabling the codes to be read even under difficult conditions of uneven printing, stained background, poor contrast, cut bars etc. Instead of merely detecting light absorption or reflection spikes when a detector is moved in front of the code, where the presence of a bar is indicated when a threshold is exceeded, the following method is used: a matrix image of black and white pixels is set up; the sum Tj of black pixels of each column is counted to draw up a curve Tj as a function of the abscissa j of columns; a threshold SDA is established for Tj, below which, in principle, the columns do not contain a portion of a bar code bar; blocks of columns going beyond the threshold SDA are demarcated, and all the other columns of the image are eliminated; then shape recognition is carried out on the remaining blocks to ascertain whether the block truly corresponds to a bar or not.