Wilson statistics, described in Chapter 2, aims at calculating the distribution of the structure factor P(F) ≡ P(|F|, φ) when nothing is known about the structure; the positivity and atomicity of the electron density (both promoted by the positive nature of the atomic scattering factors fj) are the only necessary assumptions. Wilson results may be synthesized as follows: . . . the modulus R = |E| is distributed according to equations (2.7) or (2.8), while no prevision is possible about φ, which is distributed with constant probability 1/(2π). . . . In other words, knowledge of the R moduli does not provide information about a phase; this agrees well with Section 3.3, according to which experimental data only allow an estimate of s.i. (and also s.s. if the algebraic form of the symmetry operators has been fixed). Let us now consider P(Fh1 , Fh2 ) ≡ P(|Fh1 |, |Fh2 |, φh1 , φh2 ), the joint probability distribution function of two structure factors. If the two structure factors are uncorrelated (i.e. no relation is expected between their moduli and between their phases), P will coincide with the product of two Wilson distributions (2.7) or (2.8), say, . . . P(Fh1 , Fh2 ) ≡ P(|Fh1 |, φh1 ) · P(|Fh2 |, φh2 ) = 1/4π2 P(|Fh1 |)P(|Fh2 |), . . . which is useless (because the two Wilson distributions are useless) for solving the phase problem; indeed, the relation does not provide any phase information. The question is now: if two structure factors are correlated, may their joint probability distribution function be used for solving the phase problem? Let us first use a simple example to show how much additional information (i.e. that is not present in the two elementary distributions) may be stored in a joint probability distribution function; then we will answer the question. Let us suppose that the human population of a village has been submitted to statistical analysis to define how weight and height are distributed.