In this paper the use of neural networks (NNs) to forecast daily hourly values of the ionospheric F2 layer critical frequency, foF2, at any target geographic location up to 5 hours ahead is illustrated. The inputs used for the NN are universal time, day of the year, a 2 month running mean sunspot number (R2), a 2 day running mean of the 3 hour planetary magnetic ap index (A16), solar zenith angle, geographical latitude, magnetic dip angle, angle of magnetic declination, angle of meridian relative to subsolar point, and the four recent past observations of foF2 (F−3, F−2, F−1, and F0) from that target geographic location. The outputs of the NN are F+1, F+2,F+3, F+4, and F+5, representing the values of foF2 up to 5 hours ahead of F0. In this work, data from 40 worldwide ionospheric stations spanning the period 1964–1986, which include all periods of calm and disturbed magnetic activity, were used for training the NN. In order to test the predictive ability of the NN, the NN was verified with data from 10 stations not included in the training set that were selected for their remoteness from the trained stations. The results obtained from the NN are compared with the observed values of foF2 obtained from these selected verification stations. The performance of the NN is measured by calculating the root-mean-square (RMS) error difference between the NN model and measured station data. This paper illustrates that short-term predictions of foF2 are much improved by including past observations of foF2 itself, in addition to those temporal and spatial inputs mentioned above, and that NNs can successfully be applied to the task of global forecasting.