A new measure, the coefficient of constraint, is proposed which describes the additional information in the second of the pair of letters compared with the information which that second letter might contain if it occurred alone. When calculated for a passage of printed English, a quite regular behavior of the coefficient was discovered and described by the expression D(n) = 1/n-super(2). A possible use of the coefficient is proposed by which an upper bound is set on the average information per letter in the text examined bearing close resemblance and some striking differences with Shannon's estimate of the same function. (PsycINFO Database Record (c) 2006 APA, all rights reserved)