The potential of Vis-NIR for the non-invasive determination of physicochemical attributes of dairy powders was investigated for fast assessment of dairy powder quality. Three scanning techniques were compared for the collection of Vis-NIR data, including a portable, a benchtop and a hyperspectral imaging device. Models were developed with commercial dairy powders varying in batch, brand, type and length of storage and tested in a completely independent dataset. Moderate to good performance was achieved for tapped density, insolubility index, surface free fat, moisture content and bulk density (RP2: 0.65–0.88, 0.80–0.85, 0.77–0.87, 0.71–0.86 and 0.71–0.72, respectively). Analysis using variable of importance in projection and selectivity ratio show the uniqueness of each prediction model for the different physicochemical attributes. Three scanning techniques investigated present different opportunities for on-line implementation of this system.