Numerous regionalisation studies have been conducted to predict the runoff time series in ungauged catchments. However, there are few studies investigating their benefits for predicting runoff time series on a continental scale. This study uses four regionalisation approaches, including spatial proximity (SP), gridded SP, integrated similarity (IS) and gridded IS, to regionalise two rainfall–runoff models (SIMHYD and Xinanjiang) for 605 unregulated catchments distributed across Australia. The SP and IS approaches are used for directly predicting catchment streamflow; the gridded SP and gridded IS approaches are used for predicting runoff at each 0.05° × 0.05° grid cell for continental Australia, which is then aggregated for each catchment. The IS and gridded IS approaches use five properties to build similarity indices, including three physical properties (an aridity index, a fraction of forest ratio and the mean annual air temperature) and two rainfall indices (rainfall seasonality and the standard deviation of daily rainfall). The two rainfall–runoff models show consistent regionalisation results, and there is a marginal difference among the four regionalisation approaches in the wet and densely located catchments. However, the gridded IS approach outperforms the other three in the dry and sparsely located catchments, and it overcomes the unnatural tessellated effect obtained from the gridded SP approach. Use of the gridded IS approach together with rainfall–runoff modelling for predicting runoff on a continental scale is highly recommended. Extra predictors should be included to build similarity indices in other regions, such as the high latitude northern hemisphere or high elevation regions.