This paper deals with the estimation of a gramian-based interaction measure from logged process data, and thereby removing the need of creating parametric models prior to the selection of the significant input-output interconnections. Moreover, the resulting confidence regions of the estimates can be used to perform a robust control structure selection.