Efficient gene transfer is a major challenge for non-viral gene therapy. Understanding how non-viral vectors initiate gene expression could lead to the development of new future vectors with enhanced efficacy. Linear or branched polyethylenimine (PEI)/DNA complexes were generated in varying salt conditions and their transfection efficiencies were compared in vitro and in vivo using reporter genes, luciferase and green fluorescent protein, and rhodamine labeled DNA (pGeneGrip). The transfection efficiency of linear PEI22/DNA in vitro was generally greater than that of branched PEI/DNA when complexes were generated in salt containing buffer. However, PEI complexes generated under salt-free conditions generally had low transfection activity in vitro. In contrast, PEI22/DNA salt-free complexes were highly active in vivo. Branched PEI/DNA and salt containing PEI22/DNA complexes were generally 10-100-fold less active than the salt-free PEI22/DNA complexes. Salt-free PEI22/DNA complexes were small, but subsequently grew into aggregates when salt was added. In contrast, PEI25/DNA complexes remained small even after salt was added under the same conditions. Furthermore, PEI22/pGeneGrips complexes formed large aggregates associated with the cell membrane, cytoplasm and nucleus, while branched PEI complexes remained as small distinct particles associated with the cell membrane or in the cytoplasm. Branched and linear PEI/DNA complexes differ in their ability to transfect cells. The greater efficiency of linear PEI might be due to an inherent kinetic instability under salt conditions. Understanding how to employ this kinetic instability of linear PEI could help in designing future vectors with greater flexibility and transfection efficiency in vivo.