Water stored deep in the soil profile is valuable to crop yield but its availability and conversion to grain vary with preceding management and seasonal rainfall distribution. We investigated the value of subsoil water to wheat on the Red Kandosol soils in southern New South Wales, Australia, using the APSIM Wheat model, carefully validated for the study area. Simulation treatments over 106 years of historic climate data involved a factorial combination of (1) a preceding crop of either lucerne (Dry treatment) or a low-yielding wheat crop (Wet treatment) and (2) restriction of wheat root depth to either 1.2 or 1.8 m. Root access to the subsoil (1.2–1.8 m) increased wheat yield by an average of 0.6 and 0.3 t/ha for the Wet and Dry treatments, respectively, at Cootamundra (mean annual rainfall 624 mm) and by 0.5 and 0.1 t/ha at Ardlethan (mean annual rainfall 484 mm). The differences were principally related to the frequency with which the subsoil failed to wet up, which occurred in 8% and 39% of years at Cootamundra in Wet and Dry treatments, respectively, but in 21% and 79% of years at Ardlethan. In seasons where water from the subsoil was used, the mean value of the water for grain yield, expressed as marginal water-use efficiency (MWUE), was 30–36 kg/ha.mm at both sites. High MWUE (>60 kg/ha.mm) generally occurred in seasons of above-average rainfall when subsoil water facilitated extra post-anthesis water extraction, including that from upper soil layers, to realise the high yield potential. Low MWUE (<10 kg/ha.mm) occurred when re-translocation of pre-anthesis assimilate to grain in the 1.2 m treatment compensated for reduced subsoil water extraction and no yield difference between 1.2 and 1.8 m treatments was observed. Counter-intuitively, the results suggest that subsoil water will be of more value in higher rainfall environments due to its more frequent occurrence, and in above-average seasons due to more efficient conversion to grain.