[Purpose] In this study, we show the practice and results of the sign language headword selection in the compilation of the <Korean Sign Language-Korean Dictionary>. [Method] Although this dictionary is a corpus-based dictionary, the size of the corpus is not sufficient, so for the headword selection conducted from 2018 to 2023, existing widely used sign language dictionaries and textbooks were also used to select headwords. The two types of data went through a preprocessing process to become appropriate data for the headword selection. While these materials were used as basic data, the authors also looked at the headwords for each type and added headwords if necessary. This process was carried out according to the principles of headword selection. [Results] The headwords which were selected in three ways were added together and duplicate heandwords were deleted, variants were grouped, and homonyms or polysemes were distinguished. In addition, the list of headwords was refined by adding homonyms, synonyms, and antonyms related to the selected headwords. Finally, the final list was finalized after review by the Korean Language Institute. There are a total of 2,484 headwords, and they are organized into 16 semantic categories. The headword list include 2,052 headwords selected from the Korean Sign Language corpus, and 432 headwords were selected from sign language dictionaries or textbooks and added manually by the dictionary compilers. Also a total of 344 variants were identified for the 2,484 headwords. [Conclusion] In order to compile a high-quality dictionary, it was suggested to expand the corpus, find alternative ways to secure headwords in addition to the corpus, distinguish semantic categories by fully considering the properties of polysemy, and enhance the function of the dictionary compilation system. This study is expected to contribute to the development of sign language lexicography by specifically presenting the principles and processes related to lemma selection, which have not been discussed in depth so far.