Out-of-vocabulary lexicons, including new words, collocations, as well as phrases, are the key flesh of a human language while an obstacle to machine translation. But the translation of OOV is quite difficult to obtain. A web-mining solution to the OOV translation is adopted in our research. The basic assumption lies in that most of the OOV’s translations exist on the web, and search engines can provide many web pages containing the OOV and corresponding translations. We mine the translation from returned snippets of the search engine with expanded OOV as the query term. The difference of our method from other methods lies in that a query classification is made before submitting to the search engine. Experiment shows our solution can discover the translation to many of the OOVs with quite high precision.