Conventional HF communication predictions are usually based on the observed monthly median values of ionospheric parameters. The recently acquired capability of certain users to alter their terminal parameters with very short notice can be exploited only if predictions, based on real-time observations, are available with lead times of a few hours. Such very short-term predictions of, sav, fo F2 are possible by performing autocorrelation analysis on the deviations from mean diurnal patterns of f0F2 separately for quiettime and disturbed conditions. A Markov statistical model is used to estimate persistence in the data, and hence the extent of the forcast. The forecast is shown to be better during disturbed conditions compared to the quiettime conditions. It is shown that a forecast is likely to be valid up to 3 hours under quiet conditions, and up to 4 hours and 6 hours during daytime and nighttime, respectively, under disturbed conditions. The prediction basis for these forecasts is a representative daily diurnal cycle under quiet conditions, and a diurnal pattern averaged over three diurnal cycles under disturbed conditions. Prediction methods exist (elsewhere) for the prediction of magnetic storms 3 days in advance.