Neural networks have been subject to controversy ever since the concept was first introduced in the mid-20th century. As time went by, they turned out to be a powerful tool for solving a number of machine learning tasks including pattern recognition, classification, regression and other. For some tasks, neural networks achieve significantly better results than traditional methods. On the other hand, employing neural networks in other tasks shows little or no improvement compared to the conventional methods. As regards the problem of part-of-speech tagging, numerous attempts to make use of neural networks have been made in the last few decades. A number of different configurations have been tested, from a simple perceptron trainer to multilayer feedforward networks, from pure neural networks to various mixed approaches employing other methods (e.g. Hidden Markov Models) as well. At least in some cases (e.g. Schmid's NetTagger), the neural-network-based tagger was able to level or even beat the state-of-the-art accuracy. The main goal of the thesis is to implement a neural-network-based part-of-speech tagger. This software tool will be able to read a manually tagged training corpus from which it will automatically learn rules for assigning tags to tokens; the trained tagger will in turn be able to tag an arbitrary corpus in given language. The learning algorithm should make use of the mechanisms used in the training of neural networks; nevertheless, it can combine other approaches like Hidden Markov Models or Conditional Random Fields as well. In addition to the implementation part, the thesis will also review the most notable attempts to employ neural networks in part-of-speech tagging made up to this point.