Background: Clear and detailed eligibility criteria (EC) of trial participants are necessary to apply study results to larger populations. Objective: To study the reporting of EC comparing trial protocols to subsequent publications. Methods: We established a cohort of 52 trial protocols submitted in 2000 to the Research Ethics Committee of the University of Freiburg, Germany and 78 subsequent full publications identified by electronic literature searches and survey of applicants. We extracted information on EC that differed between protocol and publication and classified them into seven content categories. For each discordant EC we examined whether it was added to the publication or omitted, and whether it was stated as inclusion or exclusion criterion. We assessed whether the difference represented a minor or major change and whether it would be suggestive of an increase or decrease of the reported study population. Results: In one trial, EC in protocol and publication were identical. For 51 trials corresponding to 77 articles we found 778 discordant EC. Of those, 70 (9%) were added to publications and 708 (91%) were missing. For 182 (23%) EC were formulated as inclusion criterion, 589 (76%) as exclusion criterion and 7 (1%) as ‘patient selection criteria’. We regarded the difference as minor for 138 (18%) EC, and major for 640 (82%). For 109 (14%) EC, the articles gave the impression of a smaller study population, for 646 (83%) of a larger one compared to protocols (23 unclear). Forty-eight per cent of EC were about patients’ co-morbidity, 22% medical treatment and 14% about disease type/severity (other 16%). We will discuss associations of trial characteristics with differences in EC. Conclusions: Authors of systematic reviews and other users of randomized evidence rely on study information as reported in journal articles. Most articles do not mirror the exact definition of study populations as documented in trial protocols.