Despite the availability of various remote sensing methods allowing for mapping, monitoring, and studying the underwater cultural heritage at previously unreachable depths, underwater operations remain costly and challenging to sustain in extensive areas. The adoption of formal models indicating where to expect archaeological remains would be extremely beneficial to optimise underwater archaeological investigations. However, whilst archaeological predictive modelling has increasingly been employed in terrestrial contexts, this technique is underdeveloped in the maritime domain, particularly in the Mediterranean basin. While hinting at a mistaken notion of what predictive models should achieve, this underdevelopment also highlights specific caveats, which should be addressed to improve current archaeological predictive modelling approaches, thus promoting their further development in maritime areas. This contribution presents a new GIS-based methodology for the prediction of shipwreck locations in Mediterranean territorial waters (i.e., 12 NM zone); particularly, it focuses on strategies to deal with data biases, model uncertainty and testing.