Data-driven learning (DDL), or corpus-based language learning, involves the learner in an exploratory task to discover appropriate expressions or collocates regarding his writing. However, the problematic units of meaning in each learner's writing are so diverse that conventional corpora often prove futile. The search engine Google with the characteristics of dynamic corpora can provide quantitative parameters for evaluating a chunk of language. This article describes an experiment in which DDL was accomplished with the aid of Google and presents the evaluation of the outcome based on a comparison of Google and the British National Corpus (BNC) and eventually reaches the conclusion that the dynamic corpus or search-engine-based corpus is superior in usability, search speed, the number of solutions and above all, preference investigations. (Contains 2 tables and 2 figures.)