Background: Observers responsible for outcome data collection in randomised clinical trials are frequently not blinded. It is often assumed that outcomes are more reliable if observers are blinded, but there is scant reliable empirical evidence on the typical degree of observer bias, and on which factors that are associated with such bias. Previous reviews have compared blinded observers in one group of trial with unblinded observers in another group of ‘similar’ trials, and risk confounding by known or unknown factors, for example concealment of allocation. Objectives: To estimate the degree of observer bias in randomised clinical trials, and to identify factors associated with large observer bias. Methods: A systematic review of randomized clinical trials with both blinded and unblinded evaluation of the same binary outcome. We searched PubMed, PsycInfo, Embase, The Cochrane Methodology Register, Google Schooler, and HighWire Press. We will compare the odds ratio for an unwanted event in each trial for the blinded and for the unblinded assessment. We define observer bias as the relative odds ratio (ROR): ORunblinded / ORblinded. We plan to summarise the RORs using random effects models. Furthermore, we will study whether the estimated observer bias is larger in outcomes involving a) large inter-observer variations (i.e. a high degree of subjective judgment, e.g. global clinical improvement), b) close observer involvement in the trial (e.g. they are clinicians that also treat patients); and c) close patient-assessor interaction (i.e. the outcome is sensitive to patient behavior). Results: At the time of abstract submission we had identified 14 eligible trials. Conclusions: Pending.