A backpropagation algorithm is used to train a neural net with the goal of distinguishing between two groups of biological species: prokaryotic and eukaryotic, based on frequencies of all 16 doublets in DNA sequences. An improvement of about 15% is obtained compared to statistical analysis based on one doublet only. This is done first by presenting sequences of species to the network with known classification (the training phase) and then showing species which the neural net has never seen before, and looking for the response. A brief discussion of the speed of training is given.