Abstract: A new king of neurons and new kind if BP networks,self learning neurons and self learning BP networks,are presented in this paper with a wiew to their implementation.A self learning BP network is composed of self learning neurons.One of the prominent characteristics of the self learning neuron is thatit has a forward channel,a backward channel and a leamer,This makes each neuron in a self learning BP network accomplish the forward propagation of message,backward propagation of errors,and modification of its parameters independently,and hence makes the network implement parallel computing easity.A new training policy for BP networks is also presented in this paper.Simulation results demonstrate the effectiveness of the policy.