This work is devoted the comparative analysis of neural networks of different 
architecture for the decision of tasks of recognition of the distorted characters 
at presence of noise of different type and intensity. The comparative analysis 
of results of recognition, got with the use of neural backpropagation network 
of reverse distribution and hierarchical neural network method, allows to recommend 
last for recognition of characters at a relation signal/noise equal 3 and presence 70% character.