The combination of corpus methodology and metaphor theory has developed into a mainstream trend in metaphor studies in recent years. The adoption of this methodology puts metaphor research on a sounder empirical footing, broadens the scope of metaphor research and to some extent addresses a number of methodological issues of some early metaphor studies. This research paradigm, however, still receives a range of doubts and challenges in the areas of data collection and identification, specifically with respect to data search, representativeness, balance and indirectness. This paper systematically categorizes and analyzes these issues and puts forward several relevant recommendations for future research.