The cytokine LIGHT activates various anti-tumor functions through its two receptors, lymphotoxin beta receptor (LTbetaR) and herpes virus entry mediator (HVEM), and is expected to be a promising candidate for cancer therapy. However, LIGHT is also trapped by decoy receptor 3 (DcR3), which is highly expressed in various tumors. Here, we used phage display technique to create LIGHT mutants that specifically bind LTbetaR and HVEM, and is not trapped by DcR3 for optimized cancer therapy. We constructed phage library displaying structural variants of LIGHT with randomized amino acid residues. After the affinity panning, we created 6 clones of LIGHT mutants as candidates for DcR3-evading LIGHT. Analysis of binding affinities showed that all candidates had 10-fold lower affinities for DcR3 than wild-type LIGHT, while 5 of the 6 clones had almost the same affinity for LTbetaR and HVEM. Furthermore, analysis of detailed binding kinetics showed that lower affinity for DcR3 is dependent on their faster off-rate. Further, we showed that the LIGHT mutant had almost the same cytotoxicity via LTbetaR, and had 62-fold higher DcR3-evading capacity compared to the wild type. Our data provide valuable information for construction of more functional LIGHT mutants that might be powerful tools for cancer therapy.