Abstract. [Context & motivation] Bidding processes are a usual requirement elicitation instrument for large IT or infrastructure projects. An organization or agency issues a Request for Proposal (RFP) and interested companies may submit compliant offers. [Problem] Such RFPs comprise natural language documents of several hundreds of pages with requirements of various kinds mixed with other information. The analysis of that huge amount of information is very time consuming and cumbersome because bidding companies should not disregard any requirement stated in the RFP. [Principal ideas/results] This research preview paper presents a first version of a classification component, OpenReq Classification Service (ORCS), which extracts requirements from RFP documents while discarding irrelevant text. ORCS is based on the use of Naïve Bayes classifiers. We have trained ORCS with 6 RFPs and then tested the component with 4 other RFPs, all of them from the railway safety domain. [Contribution] ORCS paves the way to improved productivity by reducing the manual effort needed to identify requirements from natural language RFPs