Trial registration and public accessibility of appended or published protocols of phase III randomized clinical trials (RCTs) allow comparison of reported research with essential aspects of trial design. We determined how eligibility criteria of participants specified in protocols were described in trial registries and articles of 255 cancer RCTs published in leading journals. The mean proportion of matching eligibility criteria between protocols and publications per trial (the primary endpoint) was 44.0% (95% confidence interval [CI] = 40.8% to 47.3%). Almost all discrepancies in eligibility criteria (96.7%, 95% CI = 96.1% to 97.3%) suggested to readers of articles that a broader study population was included. The mean proportion of matching eligibility criteria between protocols and registries was 72.9% (95% CI = 68.2% to 77.7%, the secondary endpoint). We conclude that there are substantial differences in eligibility criteria between trial protocols, registries and articles. Inaccurate reporting of eligibility criteria may prevent appropriate assessment of the applicability of trial results.