Conformity to observations takes on a particular significance in relation to weather prediction. Day-to-day weather changes constitute the temporal evolution of a complex and highly chaotic system, and yet the accuracy of weather forecasts can be, and are, readily and regularly assessed by both practitioners and the general public. Here, the development of weather prediction is portrayed against the backcloth of the role played by the criterion of ‘conformity to observations’ in regulating progress. It has served as a crude banner for castigating an early attempt at forecasting, sustaining an empirically-based but theoretically unsupported approach, summarily dismissing some attempts and yet ignored in another. In more recent developments the criterion has been both deliberately and justifiably weakened, and even turned on its head.