The maximum entropy principle from information theory was used to determine the six partial distribution functions of a glass in the AgGeSe system from anomalous X-ray scattering data. The distribution functions had the maximum possible entropy consistent with fitting the scattering data to within experimental precision. The maximum entropy formulation permitted controlling the degree to which experimental error influenced the results. Partial distribution functions that satisfied the data exactly were extremely sensitive to error. The scattering data consisted of scans at two energies near the K absorption edge of each glass component, for a total of six independent patterns. As a preliminary test of the method, the Ge, Se, and Ag difference distribution functions were obtained from pairs of patterns at the respective absorption edges. When highly precise data were available, the maximum entropy results closely matched those from the so-called difference approach. The maximum entropy partial distribution functions, although somewhat distorted by systematic error, provided direct evidence for structural correlations that had been surmised previously from the difference approach functions. The exact solution for the partial distribution functions, on the other hand, was completely dominated by the effects of random error.