<p>Establishing a highly efficient and clear mapping between the microstructure and macroscopic mechanical properties of dual-phase steel constitutes a significant challenge. This study presents an integrated computational approach combining finite element simulation with deep learning to establish bidirectional microstructure–property relationships for static tensile behavior prediction. A highly parameterized microstructure model of dual-phase steel is established based on the Voronoi model and the theory of random probability distribution. The validity of the model is verified through the comparison of quasi-static tensile test results with the simulation outcomes. A static tensile performance prediction model of dual-phase steel based on the Batch Normalization Residual Visual Geometry Group 16 (BN-ResVGG16) is established. A microstructure design approach of dual-phase steel driven by static tensile performance is put forward. A microstructure image generation model of dual-phase steel based on conditional adversarial generative network (cGAN) is devised. The research results show that the proposed BN-ResVGG16 prediction model realizes high-precision predictions of key performance indicators such as yield strength, tensile strength, and fracture strain of dual-phase steel. The proposed microstructure generation model based on cGAN enables the generation of microstructure images that conform to the target static mechanical properties. A bidirectional high-precision mapping between the microstructure and static tensile properties of dual-phase steel has been accomplished, offering a new solution for the application of dual-phase steel in the lightweight design of vehicle body structures and the development of high-performance materials.</p>