EGFR-targeted inhibitors (gefitinib and erlotinib) provided an effective strategy for the treatment of non-small-cell lung cancer. However, the EGFR T790M secondary mutation has become a leading cause of clinically acquired resistance to these agents. Herein, on the basis of the previously reported irreversible EGFR inhibitor (compound 9), we present a structure-based design approach, which is rationalized via analyzing its binding model and comparing the differences of gatekeeper pocket between the T790M mutant and wild-type (WT) EGFR kinases. Guided by these results, a novel 6,7-dioxo-6,7-dihydropteridine scaffold was discovered and hydrophobic modifications at N5-position were conducted to strengthen nonpolar contacts and improve mutant selectivity over EGFR(WT). Finally, the most representative compound 17d was identified. This work demonstrates the power of structure-based strategy in discovering lead compounds and provides molecular insights into the selectivity of EGFR(L858R/T790M) over EGFR(WT), which may play an important role in designing new classes of mutant-selective EGFR inhibitors.