ABSTRACTHourly observations of stream discharge and suspended sediment concentration obtained during parts of 1978 and 1981 from the proglacial stream of the glacier de Tsidjiore Nouve, Valais, Switzerland, are used to investigate the usefulness of suspended sediment rating curves. Residuals from standard ordinary least squares (OLS) rating models fitted to the data from both ablation seasons are highly serially correlated. Discriminant ratings produced by fitting standard OLS models to data from partial periods, and to data filtered according to diurnal and seasonal fluctuations in stage, climatic conditions and process lag are similarly autocorrelated. Estimates of suspended sediment concentration provided by the rating models are therefore likely to be unreliable. Forecasts from the 1981 rating models considerably underestimated observed suspended sediment concentrations, and provided only a 79% prediction of observed load in the eight days following the calibration period. Forecasts for this period from the 1978 rating models were almost as good as those from the 1981 models. In general, considerable inaccuracies occur in forecasting from rating models. In certain circumstances, however, it may be‘acceptable’to use a general rating model to forecast suspended sediment concentrations and loads for any given period. Some pragmatic ground rales for the use of rating curves are outlined.