This study deals with the Tiagba lagoon bay eutrophication modeling by using Artificial Neural Networks (ANN), principally by using Multilayer Back-propagation based on Levenberg-Marquardt algorithm. 10 or 11 inputs variables, temperature (T), pH, dissolved oxygen (DO), NH4+, NO3-and PO43-, monthly precipitation (MP), monthly debit of river Bandama (DBan), suspended matters (SM), transparence (Trans), date of sampling (DS) were used respectively for static and dynamic modeling, while one output variable (chlorophyll a) was considered for the both case. For optimization of the ANN, it was shown that the architecture 10-11-1 was the most suitable for the static modeling of chlorophyll a while the 11-11-1 architecture was the best for dynamic modeling. The validation of these models was performed by analyzing the residues. It was found that theses residues were well distributed between 0.3 and 0.4% and followed a normal law according the Henry representation in the both case. So the models obtained were suitable for the prediction of the chlorophyll a an evolution in relation of the above variables.