The Bradford law is explored theoretically by means of a very mixed Poisson model which, it is claimed, elucidates the uncertainties surrounding the law and its applications. It is argued that Bradford succeeded in formulating an empirical regularity which has pure and hybrid forms but that all the variants can be subsumed under a simple logarithmic law which, for reasons explained, escapes exact expression in conventional frequency terms. The theoretical aspects discussed include the hybridity of form, estimations, sampling problems, the stability of ranks, homogeneity of data, and tests of significance. Some numerical examples, some simulated and some drawn from social contexts outside bibliography, are used both to illustrate theoretical issues and also to indicate the wide generality of the Bradford law. Possible applications and developments of the theory are indicated.