Spatial variability of rainfall has been recognised as an important factor controlling the hydrological response of catchments. However, gauged daily rainfall data are often available at scattered locations over the catchments. This paper looks into how to capitalise on the spatial structure of radar rainfall data for improving kriging interpolation of limited gauge data over catchments at the 1-km2 grid scale, using for the case study 117 gauged stations within the 128 km × 128 km region of the Mt Stapylton weather radar field (near Brisbane, Australia). Correlograms were developed using a Fast Fourier Transform method on the Gaussianised radar and gauged data. It is observed that the correlograms vary from day to day and display significant anisotropy. For the radar data, locally varying anisotropy (LVA) was examined by developing the correlogram centred on each pixel and for different radial distances. Cross-validation was carried out using the empirical correlogram tables, as well as different fitting strategies of a two-dimensional exponential distribution for both the gauged and the radar data. The results indicate that the correlograms based on the radar data outperform the gauged ones as judged by statistical measures including root mean square error, mean bias, mean absolute bias, mean standard deviation and mean inter-quartile range. While the radar data display significant LVA, it was observed that LVA did not significantly improve the estimates compared with the global anisotropy. This was also confirmed by conditional simulation of 120 rainfields using different options of correlogram development.