Flux balance analysis (FBA) has been widely used in calculating steady-state flux distributions that provide important information for metabolic engineering. Several thermodynamics-based methods, for example, quantitative assignment of reaction directionality and energy balance analysis have been developed to improve the prediction accuracy of FBA. However, these methods can only generate a thermodynamically feasible range, rather than the most thermodynamically favorable solution. We therefore developed a novel optimization method termed as thermodynamic optimum searching (TOS) to calculate the thermodynamically optimal solution, based on the second law of thermodynamics, the minimum magnitude of the Gibbs free energy change and the maximum entropy production principle (MEPP). Then, TOS was applied to five physiological conditions of Escherichia coli to evaluate its effectiveness. The resulting prediction accuracy was found significantly improved (10.7-48.5%) by comparing with the (13)C-fluxome data, indicating that TOS can be considered an advanced calculation and prediction tool in metabolic engineering.