Objective: To explore efficacy and safety of simulated artificial pancreas in modulating stress hyperglycemia in critically ill patients., Methods: A prospective randomized controlled study was performed. Seventy-two critically ill patients with stress hyperglycemia, aged 18-85 years, acute physiology and chronic health evaluation II (APACHE II) score over 15, two consecutive random blood glucose 11.1 mmol/L or higher, glycated hemoglobin (HbA 1C ) below 0.065, unable to eat food for 3 days after inclusion, or only accepting parenteral nutrition, admitted to intensive care unit (ICU) in Shanghai Punan Hospital of Pudong New District from January 1st, 2015 to June 30th, 2017 were enrolled. The patients were divided into three groups according to the random number table method, high-intensity group and low-intensity group were injected Novolin R (high-intensity group 2/3 dosage, low-intensity group 1/3 dosage) to modulate stress hyperglycemia by simulated artificial pancreas. Simulated artificial pancreas consisted of Guardian real time glucose monitoring system (GRT system), close-circle control algorithm and micro-pump; subcutaneous injection of Humulin 70/30 was applied to modulate stress hyperglycemia in humulin group. Real-time glucose levels of interstitial fluid in abdominal wall, equivalent to blood glucose levels, 10 minutes each time, were monitored by using of GRT system for all patients in three groups. Fasting serum levels of stress hormones including epinephrine and cortisol and insulin resistance index (IRI) were recorded within 24 hours after inclusion. Mean blood glucose, blood glucose variation coefficient, blood glucose target-reaching rate, blood glucose target-reaching time, hypoglycemia rate and 6-month mortality were measured. Twenty healthy adults from health administration department of the hospital were recruited as healthy control group., Results: A total of 60 eligible critically ill patients were included in this study, each group with 20 patients. There was no significant difference in gender, age, APACHE II scores among three groups. The levels of serum epinephrine, cortisol and IRI within 24 hours after inclusion in the three groups were significantly higher than those in healthy control group. The mean blood glucose levels of humulin group, low-intensity group, high-intensity group were decreased (mmol/L: 10.2±3.2, 8.4±2.6, 8.1±2.2), the blood glucose target-reaching rate were increased [40.2% (3 295/8 196), 71.1% (5 393/7 585), 80.4% (6 286/7 818)], the blood glucose target-reaching time were shortened (hours: 49.1±5.8, 24.6±4.6, 17.5±4.2), the hypoglycemia rates were increased respectively [1.3% (108/8 196), 2.8% (211/7 585), 4.0% (313/7 818)], with statistically significant differences (all P = 0.000). There was no significant difference in blood glucose variation coefficient and 6-month mortality among three groups [blood glucose variation coefficient: (29.4±3.7)%, (28.5±5.3)%, (26.1±4.6)%, 6-month mortality: 55.0%, 45.0%, 40.0%, all P > 0.05]., Conclusions: Simulated artificial pancreas could effectively and safely modulate stress hyperglycemia in critically ill patients, high-intensity modulation could bring about better efficacy in the regulation of hyperglycemia. High-frequency blood glucose monitoring by using GRT system could promptly identify hypoglycemia and help it to be corrected.