OBJECTIVES: In a systematic review of literature, reviewing papers twice by two reviewers improves the likelihood of correctly including/excluding citations. Also, reviewing first the abstract then the full text allows the reviewer to have had more exposure to the literature, and thus to be more accurate in his decision. A statistical model was fitted and paired t-tests were performed to determine between-reviewer and between-review reliability and variation. METHODS: Inclusion/exclusion decisions made by two reviewers at the abstract and the full-text reviews of six recently conducted systematic reviews (one economic and five clinical) were analysed and compared to the final inclusion/exclusion decision. For both reviewers, sensitivity (the proportion of correctly included citations) and specificity (the proportion of correctly excluded citations) were modelled using bayesian Poisson regression. Paired t-tests were performed to evaluate the improvements of sensitivity and specificity between the abstract and the full-text reviews. RESULTS: The sensitivity of reviewer two was significantly higher in one out of six systematic reviews at abstract review, and in another one at full-text review. Reviewer two's specificity was also significantly higher in three systematic reviews at abstract review, and in one systematic review at full-text review. Reviewers were on average 89.6% sensitive and 95.7% specific at including or excluding publications at abstract review. At full-text review, the average sensitivity increased significantly (97.1%, p=0.006) whereas the specificity remained similar (95.9%, p=0.81). CONCLUSIONS: While first and second reviewers tend to have similar sensitivity in including citations at abstract review, reviewer two tends to be more accurate at excluding citations, thus increasing the likelihood of correctly excluding citations. At full-text review, sensitivity and specificity tend to be similar between the two reviewers, but sensitivity increases significantly compared to abstract review. This shows that the full-text review provides further confirmation that the included citations are indeed relevant.