This paper is based on the 2012 Penman Lecture delivered at the 11th National Symposium of the British Hydrological Society. The title is taken from a 1965 interview by Howard Penman when hydrological modelling was just really starting. In the period since then, the idea that we might easily move towards ‘physically based’ representations of hydrological processes has proven problematic. It is argued that this might best be done within a hypotheses testing framework, where the hypotheses are the model representations of processes over some discrete elements of a catchment, integrating the small-scale variability within those elements. This might still require some distribution function to reflect that variability, since the extremes of the variability might be important in controlling the response. Hypotheses will need to be formulated that reflect the interaction of the water flow pathways and the biota. Testing of those hypotheses will require a proper account of the uncertainties inherent in the study of hydrological systems, including a recognition that many sources of uncertainty are epistemic in nature rather than simple random variability. Such uncertainties will only be significantly reduced by the development of new measurement techniques that provide useful information at the element scales of interest.