The on-line characterisation of minerals, and an ability to use this information to perform on-line sorting, opens up new opportunities for the mining industry to both improve their existing operations and exploit presently uneconomic mineral reserves. In the present study, the dielectric properties of iron ore samples have been determined over the 0.7-20 GHz frequency range, using a single microwave probe, and the key features of the measured spectra extracted using principal components analysis. The subsequent sorting of the samples, on the basis of the identified spectral features, is then automatically performed using an ANN based classification scheme. The technique has been demonstrated to successfully classify iron ore samples with minor differences in composition into ore groups that relate petrological features to metallurgical performance.