Growing evidence suggests that glycaemic variability increases diabetic complications. However, the significance of glycaemic variability in critically ill patients remains unclear. We evaluated the predictors of glycaemic fluctuation and its association with critical care outcomes. This is a nested-cohort study within a clinical trial in which 523 patients at a medical surgical intensive care unit were randomised to either intensive insulin therapy (target glycaemic control: 4.4 to 6.1 mmol/l) or conventional insulin therapy (target control: 10.0 to 11.1 mmol/l). Glycaemic fluctuation was defined as the mean difference between the highest and lowest daily blood glucose. Patients were divided into wide and narrow fluctuation groups according to the median glycaemic fluctuation (6.0 mmol/l). The association between glycaemic fluctuation and different intensive care unit outcomes was studied. Predictors of glycaemic fluctuation were age (odds ratio for each year increment 1.03, 95% confidence interval 1.02 to 1.05), diabetes mellitus (odds ratio 3.00, 95% confidence interval 1.74 to 5.17), and daily insulin dose (odds ratio for each unit increment 1.04, 95% confidence interval 1.03 to 1.05). Similar levels of glucose fluctuation were observed in intensive insulin therapy and conventional insulin therapy patients. Wide glycaemic fluctuation was associated with higher mortality (22.2 vs. 8.4%, P < 0.001). Glycaemic fluctuation was identified as an independent predictor of intensive care unit mortality (odds ratio per mmol 1.08, 95% confidence interval 1.00 to 1.18) and hospital mortality (odds ratio per mmol 1.09, 95% confidence interval 1.02 to 1.17) using multivariate logistic regression analysis. In conclusion, wide glycaemic fluctuation is an independent predictor of mortality in critically ill patients. Whether reducing glycaemic fluctuation would lead to better outcomes needs further evaluation.