This paper provides a brief overview on document analysis and recognition area, highlighting main steps and modules that are used to build recognition systems of the mentioned type. We underline basic workflow of such system down to the problem of single character recognition problem and highlighting possibilities and ways for artificial neural networks usage. Further we are conductinga formal comparison of abilities of printed characters recognition between two well known types of second generation neural networks, namely feed-forward back-propagation multilayer perceptron (MLP) and Kohonen self-organizing features map (SOM).