Tides are usually predicted by harmonic analysis, which is a superposition of many sinusoidal constituents with amplitudes and frequencies determined from an analysis of locally measured sea levels. To form accurate tidal predictions using the method of harmonic analysis, long-term sea level registrations have to be collected and analysed. This article presents an alternative approach to the task of tidal forecasts based on the application of harmonic analysis and back-propagation, artificial neural networks to short-term measurements. Sea level records from the Hillarys Boat Harbour tide gauge, Western Australia, are used to implement this original methodology and to test its performance. The results of the methodology validation show that short-term sea level registrations can be effi-ciently employed to produce accurate tidal predictions.