A structural health monitoring system based on Bayesian damage classification and DNA expression data is studied in this paper. Transplanted from the DNA array concept in molecular biology, the proposed structural health monitoring system is constructed by utilizing a double-tier AR-ARX regression process to extract the expression array from the structural time history recorded during external excitations. The AR-ARX array is symbolized as the various genes of the structure in the viewpoint of molecular biology to reflect the possible damage condition existing in the structure. A scale-down six-story steel building located at the shaking table of the National Center for Research on Earthquake Engineering was used as the benchmark structure, and the structural response with different damage levels and locations under ambient vibration was collected to support the database for structural health monitoring. To improve the feasibility of the proposed structural health monitoring system in practical application, the system is upgraded again using the likelihood selection method. The AR-ARX array representing the DNA array of the health condition of the structure is first evaluated and ranked. Totally 30 groups of expression array are regenerated from the combination of six damage conditions.. To keep the coefficient number unchanged, the best four coefficients among every expression array are selected to form the optimized structural health monitoring system, and the sequence of the array coefficients is assembled based on the likelihood score calculated for each coefficient. Test results from ambient showed that comparing to the previous damage classification system, the detection accuracy of structural damage can be enhanced by the optimized AR-ARX array perfectly. The feasibility of transplanting the DNA array concept from molecular biology into the field of structural health monitoring has been demonstrated by the proposed SHM system.