Modelling net primary productivity (NPP) is an important instrument for analysing carbon exchange between atmosphere and vegetation as well as for quantification of carbon sinks and sources. Remote-sensing-based models allow for regional NPP estimation and are potentially transferable to new regions. Comparative model analyses, however, are lacking, especially for semi-arid environments. In this study, two recent remote-sensing-based NPP models were applied for the first time to a study region in semi-arid Kazakhstan: RBM, a light-use-efficiency model based on MODIS products, and BETHY/DLR, a soil-vegetation-atmosphere-transfer model. Differences in intermediate products, their influence on calculated NPP, as well as output products are evaluated and discussed. BETHY/DLR calculates higher NPP (mean annual NPP 2010 and 2011: 136.87 g C m-2 and 106.69 g C m-2) than RBM (62.14 g C m-2 and 54.61 g C m-2) and shows stronger inter-annual changes. Spatial and seasonal patterns present well phenological differences. Comparison to field data from 2011 showed better results for BETHY/DLR, though both results were highly correlated to the field observations (BETHY/DLR: R2=0.95, RMSE=8.36 g C m-2; RBM: R2=0.98, RMSE=22.49 g C m-2). The parameterization of the light use efficiency is critical for RBM; also MODIS based 16-day time steps might be too long to capture variable climatic conditions. For BETHY/DLR, the MODIS land cover product applied in this study differentiates insufficient classes within the semi-arid environment; a more detailed land cover map is needed to improve the regional analysis.