The spatial distribution and temporal variation of primary production (PP) in the South China Sea (SCS), one of the largest marginal seas, is far from clear. In this study, to overcome the limited observations from ship-based measurements, PP of SCS is estimated by three mainstream models to obtain long-term PP time series using data from ocean color satellites. Among the three models, the absorption-based model (AbPM) showed better performance for waters depth deeper than 200 m (R² = 0.53, N = 39), where the uncertainties of satellite estimated optical and biological properties are lower. In comparison, the R² value for a chlorophyll-based model (VGPM) was 0.35 (N = 39). Furthermore, we observed that the PP spatial patterns of the three models were similar, but disagree on the annual PP magnitudes. For example, for basin waters, the spatial deviation of VGPM and CbPM compared to AbPM is up to 50%. More importantly, it is found that the long-term trend of PP is model-dependent. For instance, for basin waters the AbPM results show a prominent decreasing trend (-25.1% per decade, p < 0.05) for the period of 2003 to 2012, but it is not significant for PP results from the VGPM (-5.2% per decade, p = 0.48). The Pacific Decadal Oscillation (PDO) is found fitting well with the decreasing trend of PP revealed by AbPM, supporting the plausibility of the AbPM trend. These results further highlight the importance of a robust model in accurately capturing large-scale spatiotemporal dynamics of primary production in marine environments.