In this paper, Bayesian network that learns conditional probabilities by 
neural networks is introduced. Bayesian network is a probabilistic model
for probabilistic reasoning in the AI context. 
In order to apply this model to a real world problem which includes 
non-linear, multi-dimensional and continuous random variable, we use neural
networks as flexible representations of conditional probabilities of Bayesian
network. The model is constructed and learnt from samples collected in the real
environment. Learning conditional probability is realized by the back 
propagation algorithm. However, effective constructing graph structure of 
Bayesian network is still open problem. For experimental evaluation of graph
learning, we develop Bayesian network software, "BAYONET". This simulator
can connect to major database software, so practical large database can be 
tractable. The latter half of this article explains this software for 
structure learning of Bayesian network.