[Context and Motivation] Content-based recommender systems for requirements are typically built on the assumption that similar requirements can be used as proxies to retrieve similar software. When a new requirement is proposed by a stakeholder, natural language processing (NLP)-based similarity metrics can be exploited to retrieve existing requirements, and in turn identify previously developed code. [Question/problem] Several NLP approaches for similarity computation are available, and there is little empirical evidence on the adoption of an effective technique in recommender systems specifically oriented to requirements-based code reuse. [Principal ideas/results] This study compares different state-of-the-art NLP approaches and correlates the similarity among requirements with the similarity of their source code. The evaluation is conducted on real-world requirements from two industrial projects in the railway domain. Results show that requirements similarity computed with the traditional tf-idf approach has the highest correlation with the actual software similarity in the considered context. Furthermore, results indicate a moderate positive correlation with Spearman’s rank correlation coefficient of more than 0.5. [Contribution] Our work is among the first ones to explore the relationship between requirements similarity and software similarity. In addition, we also identify a suitable approach for computing requirements similarity that reflects software similarity well in an industrial context. This can be useful not only in recommender systems but also in other requirements engineering tasks in which similarity computation is relevant, such as tracing and categorization.