In this paper we follow Marschak in exploring the structure of Blackwe
ll's equivalence classes of information systems. The main result is a
full characterization theorem showing that two information systems are
equally informative iff for every given a priori probability vector b
oth generate identical sets of posterior probability vectors and some
technical condition is met. We then show that for some special subsets
of information systems, two information systems are equally informati
ve iff one is a permutation of the other.