Subspace based identification of model parameters requires a low rank data model. To obtain such a model it is common to collect several snapshots of the observed vector valued sequence in a larger vector. In many cases, this “window” procedure results in a new vector valued process which can be viewed as originating from a low rank data model. When studying the performance of weighted subspace based techniques this window procedure complicates both the analysis and the implementation. In this paper an attempt is made to clarify the theoretical and numerical difficulties when applying weighting in such scenarios