In this paper we proposed an evolutionary approach for answering open-domain factoid questions, which include searching among sentences that are candidate for the final answer with Memetic Algorithm (MA), and using lexical and syntactic features for calculating fitness of the sentences. Our main purpose is making a search engine with accurate answering ability, or a web-based Question Answering (QA) system. The Text Retrieval Conference (TREC) QA Tracks data are used to develop and evaluate the approach. The answering process begins with retrieving related documents from a search engine. Then, MA searches among all the sentences of these documents and finds the best one. Finally, one or more words will be extracted based on our hand-made patterns. The results of different approaches for local search, mutation, and crossover, and also different values for number of reproduction and retrieved documents are investigated in the empirical study section. The results are promising with sufficient retrieved documents, and we have obtained a threshold value for this variable. Using MA instead of examining all the sentences is a trade-off between lowering the process time and sacrificing the accuracy, but the results show that the Mametic-based approach is more efficient.