The development of a new dynamical seasonal prediction model, ACCESS-S, to replace the Predictive Ocean-Atmosphere Model for Australia (POAMA2), signifies a major achievement in addressing knowledge gaps in seasonal climate prediction. For northern Australia, it is expected that the ACCESS-S modelling framework will provide improved seasonal forecast skill in wet season rainfall onset prediction and associated climatic drivers that impact grazing industries, vital to the economy and exports across the north. To this end, the hindcast skill of the ACCESS-S1 is assessed for an index that describes northern Australian rainfall onset, defined as when 50 mm of precipitation has accumulated after the 1 September (Drosdowsky & Wheeler, 2014). We focus on the multi-model ensemble of 11 hindcast members over 1990-2012, which are initialised at 8-day intervals from the 1^st May to the 1^st September. Along with assessing the skill of the raw model hindcasts, re-gridded to the 5 km observational grid over Australia, we also test the skill improvement of quantile-quantile calibrated and mean bias-corrected model hindcasts. The raw ACCESS-S1 hindcasts capture the broad-scale features of the median rainfall onset, including the early October accumulation over southeast Queensland and near Darwin. The greatest improvement in the prediction of the median and variability in the onset is found in the calibrated hindcasts. This is also true for Brier skill scores, with the calibrated hindcasts outperforming the raw model and bias corrected hindcasts when assessed against observed median onset dates. Hence, based on its simulation of realistic onset dates and variability alone, ACCESS-S1 can be considered an improvement over POAMA2, and gives added value to the decision making of cattle producers and graziers across northern Australia.