Algorithms to retrieve ocean color from space, deterministic or statistical, often use a simplified water reflectance model, specified by a few parameters (e.g., chlorophyll concentration, backscattering and absorption coefficients at a given wavelength). The model, however, may not be representative of the worldwide ocean conditions, since many variables affecting reflectance are fixed at some average values. In this context, the semi-analytical model of Park and Ruddick (2005), PR05, used in the spectral matching POLYMER algorithm (Steinmetz et al., 2011), is examined in terms of its ability to represent properly water reflectance. The PR05 model depends on chlorophyll-a concentration, a parameter specifying the contribution of algal and non-algal particles to the backscattering coefficient, and a parameter allowing different absorption coefficients for dissolved organic matter. Model estimates at MODIS wavelengths, obtained for a representative set of Case 1 and Case 2 waters, are compared with Hydrolight calculations that include fluorescence and Raman scattering and AERONET-OC measurements. The accuracy of retrieving inherent optical properties (IOPs) using the reconstructed reflectance is also evaluated. The model parameters that give the best fit with the simulated data are determined. Agreement is generally good between the two- or three-parameter model results and Hydrolight/AERONETOC values, even in optically complex waters, with discrepancies much smaller than typical atmospheric correction errors. Significant differences exist in some cases, but having a more intricate model (i.e., using more parameters) might not guarantee convergence of the inversion scheme. The trade-off is between efficiency/robustness and accuracy. Significant errors are observed when using the model estimates to retrieve IOPs. Importantly, the model parameters that best fit the input data, in particular chlorophyll-a concentration, may not represent adequately actual values. The reconstructed water reflectance, not the retrieved model parameters, should be used in bio-optical algorithms.