Coenzyme engineering, especially for altered coenzyme specificity, has been a research hotspot for more than a decade. In the present study, a novel computational strategy that enhances the hydrogen‐bond interaction between an enzyme and a coenzyme was developed and utilized to alter the coenzyme preference. This novel computational strategy only required the structure of the target enzyme. No other homologous enzymes were needed to achieve alteration in the coenzyme preference of a certain enzyme. Using our novel strategy, Gox2181 was reconstructed from exhibiting complete NADPH preference to exhibiting dual cofactor specificity for NADH and NADPH. Structure‐guided Gox2181 mutants were designed in silico and molecular dynamics simulations were performed to evaluate the strength of hydrogen‐bond interactions between the enzyme and the coenzyme NADPH. Three Gox2181 mutants displaying high structure stability and structural compatibility to NADH/NADPH were chosen for experimental confirmation. Among the three Gox2181 mutants, Gox2181‐Q20R&D43S showed the highest enzymatic activity by utilizing NADPH as its coenzyme, which was even better than the wild‐type enzyme. In addition, isothermal titration calorimetry analysis further verified that Gox2181‐Q20R&D43S was able to interact with NADPH but the wild‐type enzyme could not. This novel computational strategy represents an insightful approach for altering the cofactor preference of target enzymes.DatabaseModel data have been deposited in the Protein Model Database database under the accession numbers PM0079165, PM0079166, PM0079167, PM0079168 and PM0079169.