The mechanical organization of knowledge for retrieval of stored information can no longer neglect the developments of point-to-point communication theory, since both deal with information and handle it by machines. The most versatile retrieval systems are those which delegate a separate tally to each information item, and which impress marks on the tally for the machine to read and to use for selective purposes. Coding is the relationship between these marks and the intellectual content of the information items. Coding determines the complexity of the selective machine and the utility of the whole process. A set of invariant coding principles is stated which define maximum coding efficiency for any tally selecting machines, and parallels are drawn between these principles and the conclusions of modern point-to-point communication theory. Zatocoding is defined—the system which superimposes random subject code patterns on the tally—and it is found to obey each of the invariant principles of coding efficiency while still allowing the simplest possible selector machine structure.