Knowledge increase ability is of great importance in the field of artificial neural network(ANN), which is also an open problem and absorbs most research attentions. Such researches will promote the further development of ANNs both in theory and practice. The knowledge increase ability of ANNs is discussed in depth. Theoretical analysis is firstly made in view of generalization ability, which results in a most promising solution of multiple network approach. Knowledge increase can be realized via knowledge accumulation and inheritance between single network and the network system. The conception of autonomy is of great significance for knowledge increase ability of ANNs. An autonomous artificial neural network(AANN) model is introduced to avoid centralized confidence assignment, which enables distributed confidence assignment and makes the system extensible. An experimental system is built on AANN units to testify its feasibility and the results are encouraging.