4-Hydroxyphenylpyruvate dioxygenase (EC 1.13.11.27, HPPD) is one of the most widely studied herbicide targets and has gained significant attention. To identify potential effective HPPD inhibitors, a rational multistep virtual screening workflow was built, which included CBP models (based on the receptor-ligand interactions in the crystal complex), Hypogen models with activity prediction ability (according to the derivation of structure-activity relationships from a set of molecules with reported activity values), and a consensus docking procedure (consisting of LibDock, Glide, and CDOCKER). About 1 million molecules containing diketone or β-keto-enol substructures were filtered by Lipinski's rules, CBP model, and Hypogen model. A total of 12 compounds with similar docking postures were generated by consensus docking. Eventually, four molecules were screened based on the specific binding pattern and affinity of the HPPD inhibitor. The biological evaluation in vivo displayed that compounds III-1 and III-2 exhibited comparable herbicidal activity to isoxaflutole and possessed superior safety on various crops (wheat, rice, sorghum, and maize). The ADMET prediction (absorption, distribution, metabolism, excretion, and toxicity) showed that compound III possessed relatively good toxicological results. This work provides a theoretical basis and valuable reference for the virtual screening and molecular design of novel HPPD inhibition herbicides.