The article presents a comparative analysis of the use of direct search algorithms to perform automatic tuning and assessment of parameter values for two models of nonstratified aquatic ecosystems: the Neva Bay (NB) and the Vistula Lagoon (VL) of the Baltic Sea. Both models are built on the same methodological principles, but they have different structures, different numbers of model variables and empirical parameters. For the VL ecosystem, the biohydrochemical block consists of two parts: the «Plankton block» and the «Benthos» block. Modeling of the nutrients transformation processes in the lagoon was carried out jointly for the aquatic environment and the upper (active) layer of bottom sediments. The NB ecosystem model uses the «Plankton block», and all processes are described only for the aquatic environment. Modeling enables to quantitatively investigate the mechanisms of biohydrochemical transformation processes: compounds of C, N, P and Si in the ecosystem of the VL; N and P compounds in the NB ecosystem. Both models provide the means to study the features of spatiotemporal dynamics of oxygen dissolved in water. The NB ecosystem model contains 58 empirical parameters, while the VL ecosystem model contains 330 ones. The Theil criterion was used as a criterion for the models adequacy. The optimization problem is solved due to the two-stage computer-aided direct search algorithm. A random search algorithm was used at the first stage. At the second stage, a local search is performed using the modified Nelder-Mead simplex method. The used algorithms proved to be effective. They ensure reliability and computational convergence of the obtained calculation results. The article shows that interannual differences in parameter values are statistically significant for a number of parameters of the models under consideration. The models sensitivity to changes in parameters values was assessed.