Practical application of hydrological models requires parameter transfer, both temporally and spatially, to compensate for the lack of data. In this study, the transferability of parameters is evaluated using a lumped hydrological model called the Xinanjiang model to simulate runoff at different spatiotemporal scales in the Jianxi basin in south-east China and its four sub-basins. The functional relationships are built based on the posterior distribution derived by the Differential Evolution Adaptive Metropolis (DREAM) algorithm. The results show that (1) the sensitivity of parameters KE, SM, KI and KG shows obvious temporal characteristics, and the sensitivity of NK and CG shows strong spatial characteristics; (2) most relationships between sensitive parameters and scales are remarkable with goodness-of-fit coefficient higher than 0.9, which has been verified to achieve good performance in the temporal and spatial transfer; (3) the spatial transferability of the model is greatly influenced by the difference between the basin sizes; and (4) those parameters with strong spatial characteristics, such as NK and CG, show obvious impacts on the performance and uncertainty of the model transferred from the larger/smaller to smaller/larger basins.