The problem of dynamic memory organization, one of the main challenges of theoretical neurobiology, of cognitive psychology, and of computer sciences is discussed. Semantic memory is interpreted here in terms of hierarchical dynamic system theory. Both neural and mental memory architectures are considered.Important conceptual models of semantic memory at the cognitive level are based on the assumption that the knowledge structure can be represented by networks. The complexity of memory structures is analyzed by graph-theoretical concepts. Dynamic models for memory development are shortly mentioned. Neural and cognitive levels can be connected by hierarchical neural network models exhibiting capability of storing and recalling correlated patterns.