Multilayer perceptron (MLP) neural networks for regional satellite ocean color applications have been developed and assessed using in situ data from various European seas. Considered MLP products are chlorophyll a concentration (Chl–a), absorption by yellow substance at 412 nm (ays(412)) and concentration of total suspended matter (TSM), all determined from spectral remote sensing reflectance RRS. Investigated oceanographic regions comprise the Eastern Mediterranean Sea, the northern Adriatic Sea, the Western Black Sea and the Baltic Sea. The in situ measurements applied in the study were produced within the framework of the Coastal Atmosphere and Sea Time Series (CoASTS) and Bio-Optical Mapping of Marine Properties (BiOMaP) programs contributing to a unique dataset that represents different water types including chlorophyll a, yellow substance and sediment dominated waters. Performance analysis of the proposed regional MLPs indicates that Chl–a can be quantified with the highest accuracy in the Eastern Mediterranean Sea (with absolute percent difference of 14% with respect to in situ measurements). In the case of ays(412), the most accurate determination is observed for the Baltic Sea waters (13%). Instead, TSM retrieval is the most accurate in the Black Sea (14%). The study demonstrated the limited generalization capability of regional algorithms. Within this context, saturation of MLP output occurring with input data not statistically represented in the training set has been investigated through cross-basin product analysis in view of proposing a practical solution to the problem.