Abstract: Accurate estimation of diffuse radiation is of great significance for evaluating its effect on terrestrial ecosystem carbon exchange. Based on the observed diffuse radiation data in the meteorological observation field in midsubtropical Qianyanzhou, Jiangxi, China from March 1, 2012 to February 28, 2013, the simulated results of five widely used diffuse radiation decomposition models (Reindl-1, Reindl-2, Reindl-3, Boland, BRL) were validated. The results indicated that, on the 30 min scale, all of the five models could well simulate the diffuse radiation of this area overall. But the effect of models decreased significantly with the rising of clearness index (k_t). Especially when k_t>0.75, each model was unable to simulate diffuse radiation in the region. Regarding the simulation of seasonal change of diffuse radiation, the five models could simulate diffuse radiation well in most months. Relative deviation between simulated and observed values of yearly diffuse radiation of five models had a maximum of 7.1% (BRL), a minimum of 0.04% (Reindl-1), and an average of 3.6%. The simulated values of the five models appeared to be overestimated in the summer when radiation was strongest, temperature was highest, and precipitation was relatively low. For example, in July, the diffuse radiation was overestimated by 14.5%-28.2%, 21.2% on average. This was primarily due to the method of estimating diffuse radiation under the condition of high k_t. The uncertainty requires further evaluation in the model application. Considering the results of validation, simulation precision and the accessibility of input variables, the order of the simulation performance of five models was BRL>Reindl-3>Reindl-2>Reindl-1>Boland.