Introduction Using a randomized controlled trial comparing tight glucose control with a computerized decision-support systems and conventional protocols (post hoc analysis), we tested the hypothesis that hypoglycemia is associated with a poor outcome, even when controlling for initial severity. Methods We looked for moderate (2.2-3.3mmol/l) and severe (<2.2mmol/l) hypoglycemia, multiple hypoglycemic events (n[greater than or equal to]3), and the other main components of glycemic control (mean blood glucose level and blood glucose coefficient of variation (CV)). The primary end-point was 90-day mortality. We used both a multivariable analysis taking into account only variables observed at admission and a multivariable matching process (greedy matching algorithm, caliper width of 105 digit with no replacement). Results A total of 2,601 patients were analyzed and divided into three groups: no hypoglycemia (n=1,474), moderate hypoglycemia (n=874, 34%), and severe hypoglycemia (n=253, 10%). Patients with moderate or severe hypoglycemia had a poorer prognosis as shown by a higher mortality rate (36% and 54% respectively, vs. 28%) and decreased number of treatment-free days. In the multivariable analysis, severe (OR 1.50, 95% CI 1.36-1.56, P=0.043) and multiple hypoglycemic events (OR 1.76, 95% CI 1.31-3.37, P<0.001) were significantly associated with mortality whereas blood glucose CV was not. Using multivariable matching, patients with severe (53 vs. 35%, P<0.001), moderate (33 vs. 27%, P=0.029), and multiple hypoglycemic events (46 vs. 32%, P<0.001), had a higher 90-day mortality. Conclusion In a large cohort of ICU patients, severe hypoglycemia and multiple hypoglycemic events were associated with increased 90-day mortality. Trial registration Clinicaltrials.gov Identifier NCT01002482 Registered 26 Octobrer 2009.