We present a re-analysis of archival HST/NICMOS transmission spectroscopy of the exoplanet system, HD 189733, from which detections of several molecules have been claimed. As expected, we can replicate the transmission spectrum previously published when we use an identical model for the systematic effects, although the uncertainties are larger as we use a residual permutation algorithm in an effort to account for instrumental systematics. We also find that the transmission spectrum is considerably altered when slightly changing the instrument model, and conclude that the NICMOS transmission spectrum is too dependent on the method used to remove systematics to be considered a robust detection of molecular species, given that there is no physical reason to believe that the baseline flux should be modelled as a linear function of any chosen set of parameters.