The effectiveness of ecosystem restoration projects is reduced by variable and unpredictable outcomes. Previous syntheses have investigated restoration effectiveness by comparing average restored conditions to average conditions in unrestored or reference systems. Here, we provide the first quantification of the extent to which restoration affects both the mean and variability of biodiversity outcomes, through a global meta-analysis of 85 terrestrial restoration studies. We found that, relative to unrestored (degraded) sites, restoration actions increased biodiversity measures by an average of 20%, while decreasing the variability of biodiversity (quantified by the coefficient of variation) by an average of 15%. As restorations aged, mean biodiversity increased and variability in biodiversity decreased relative to unrestored sites. However, restoration sites remained, on average, 10% below the biodiversity of reference (target) ecosystems, and were characterised by a higher (15%) amount of variability. The lower mean and higher amount of variability in biodiversity at restored sites relative to reference sites remained consistent over time, suggesting that sources of variation (e.g., prior land use, restoration practices) have an enduring influence on restoration outcomes. Our results point to the need for new research to confront variability within outcomes, and close variability and biodiversity gaps between restored and reference conditions.