Drug-likeness, comprising absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties, plays a significant role in early drug discovery. However, as for current strategies of lead optimization, in vitro potency is still the focus, which may cause "molecular obesity" (poor ADMET properties). Therefore, optimization of ADMET properties would be a preferable complement for drug discovery. In this paper, we present a web server, ADMETopt, that applies scaffold hopping and ADMET screening for lead optimization. More than 50 000 unique scaffolds were extracted by fragmenting chemicals deposited in the ChEMBL and Enamine databases. Up to 15 ADMET properties can be predicted to screen the potential molecules, including seven physicochemical properties and eight biological properties. All of the models were built in terms of our previous studies and are available in our web server admetSAR. For the plausibility measurement of the modified molecules, synthetic accessibility and quantitative evaluation of drug-likeness were then implemented. As a case study, a scaffold similarity network was constructed for compounds that have bioactivities on estrogen receptors. The results demonstrated that the feasibility and practicability of our web server are acceptable. The web server is publicly accessible at http://lmmd.ecust.edu.cn/admetsar2/admetopt/ .