We present the results of the neural network (NN) investigation.
The NN contains the neural ensembles (NE) that are formed during NN training. 
NE is a set of neurons with connections among them and the connection weights 
are stronger than between other neurons. Every ensemble presents one entity, for example, 
any characteristic, a feature, an object, any relation. 
The main problem is how many ensembles we can save 
in the NN with possibility to recover them from a part of information.
We investigate the NN information capacity. We describe the experiments 
and give the results of this investigation. The main conclusion of the investigation 
is that we can save and restore the NEs and the number of these ensembles
can be more than a NN neuron number. 
This type of NNs can be used in recognition tasks 
of different situations and as an element of decision making system.