A pair of models of fractal recursion on city hierarchy, fm=f1r1-mf and Pm=P1rm-1p, are derived using entropy-maximizing methods, and the relationship of inverse proportion between the number (fm) of cities at a given level of the urban hierarchy and the average population size (Pm) of the fm cities is established, i.e fm∝1/P. It is demonstrated that the underlying rationale of both the scale law of city rank-size distribution and the Zipf dimension value in standard state (dz=lnrp/lnrf=1) rests with maximization of information entropy of city hierarchies.