Recent results deriving the star formation history of galaxies from sub-mm surveys - presented as conclusive - show a significant discrepancy with respect to previous works based on Herschel and multi-wavelength data. What is responsible for such a (puzzling) difference? Indeed the results are extremely sensitive to the template SEDs considered to derive the total IR luminosity (it is crucial to sample the dust emission peak). Here I will show how the 850-micron selection is not optimal to reconstruct the whole history of the dust obscured SFR: it might miss a large fraction of sources and, if not combined with shorter wavelength (IR) data, is not able to reliably reconstruct the total IR luminosity function.