An optimal design methodology is proposed for micro bare-tube heat exchangers. A simulated annealing method is employed with a trained neural network representing the heat transfer and pressure drop characteristics of a specified tube bank. A commercial CFD code, FLUENT5, is used to obtain the heat transfer and pressure drop data sets for in-line tube bundles, which are then used to train the neural network. Three types of micro bare-tube heat exchangers are designed and compared to conventional commercial heat exchangers. The optimized micro bare-tube heat exchangers show better performance than conventional gas-liquid heat exchangers.