Gustavsen and Semlyen’s vector fitting algorithm (VFA) is popular for the determination of the poles and residues of a sampled response function, modeled as a finite order rational function. However, it does not always perform well when the original function is of infinite order and/or the model order is overestimated. This article is concerned with a novel “null-space” modification of the VFA that provides an accurate representation of the original function when the model order is large and when the original VFA performs poorly. The null-space method is parameterized by a single control parameter, L, which can be adjusted to improve accuracy. Comparisons are made with the original VFA and another of Gustavsen’s modifications