Typesetting of research papers was way more complex (for authors) in the pre-personal computer days (pre 1990, say). So you needed publishers to pay and employ

Typesetting of research papers was way more complex (for authors) in the pre-personal computer days (pre 1990, say). So you needed publishers to pay and employ people to do this. And then you also have the logistics of mailing the journals to thousands of libraries across the world. The time and effort involved in this certainly justifies at least some of the conventional costs, given that the cost scales almost linearly with the number of subscribers, number of pages, etc. The big 'step-change' with the internet is that this linear scaling doesn't exist any more, we are reduced to a fixed cost and it is price-gouging for publishers to charge per page, per article, etc. etc. Ninja edit: In the internet era, here is one post that describes how much it costs to run a journal, the prestigious Journal of Machine Learning Research http://blogs.law.harvard.edu/pamphlet/2012/03/06/an-efficient-journal/ http://blogs.law.harvard.edu/pamphlet/2012/03/06/an-efficien...