Nearest-neighbor pair distribution functions are computed from the semiflexible chain lattice cluster theory (LCT) for binary polymer blends and are compared with the predictions of simple random mixing theory. The LCT treats lattice model polymers with structured monomers and with variable chain flexibility by allowing the monomers, to extend over several lattice sites and by introducing trans <----> gauche bending energies. Comparisons with Monte Carlo simulations for polymer melts enable further tests for the accuracy and limitations of the LCT, while computations of nearest-neighbor pair distribution functions for a variety of binary polyolefin blends provide a link between the phase behavior of these blends and the microscopic local correlations induced by packing constraints and energetic interactions. Altering monomer structures leads to subtle changes in pair distribution functions but profound variations in phase behavior. The calculations of the nearest-neighbor Fair distributions provide a simple microscopic explanation for the LCT predictions of the pressure dependence of blend phase diagrams. Variations of the nearest-neighbor pair distribution functions with chain stiffness, van der Waals interactions, temperature, polymerization indices, etc., are correlated with trends in phase behavior and other physical properties. The semiflexible chain LCT is used for further tests towards developing a computationally convenient thermodynamically equivalent linear semiflexible chain model to mimic the melt and blend properties of experimental (or theoretical structured monomer chains. (C) 1997 American Institute of Physics.