To clinically validate the precision of diagnostic Sepsis-3 criteria, and to guide and generalize its clinical application. A multicenter retrospective observational study was conducted. The patients admitted to intensive care unit (ICU) of 6 tertiary hospitals in Zhejiang Province from January to June 2015 were enrolled, and the patients satisfying the diagnostic criteria of Sepsis-2 and Sepsis-3 were screened. Population characteristics between the patients satisfying two editions were compared, and the diagnosis accuracy rate in different degree hospitals were investigated. According to the doctor's diagnosis, the patients who met the criteria of Sepsis-2 were divided into diagnosis group and non-diagnosis group, and the factors influencing the diagnosis of sepsis were analyzed by logistic regression. The patients meeting Sepsis-2 but no meeting Sepsis-3 were served as exclusion group, and those meeting Sepsis-2 and Sepsis-3 were served as enroll group, and the characteristics of patients between the two groups were compared. Receiver operating characteristic (ROC) curve was plotted to evaluate the predictive value of systemic inflammatory response syndrome (SIRS) score, sepsis-related quick sequential organ failure assessment (qSOFA) and sequential organ failure assessment (SOFA) on death, and whether the consistency of qSOFA and SOFA would affect the sensitivity of definition. The patients meeting Sepsis-2 were divided into non-survived group and survived group, and the factors associated with death were analyzed by logistic regression. Finally, 1 423 patients were enrolled, 3 patients with age < 18 years and 19 patients with missing data were excluded. There were 363 patients and 329 patients met Sepsis-2 and Sepsis-3, respectively. No significant differences were found in population characteristics between the groups of Sepsis-2 and Sepsis-3 (all P > 0.05) except for acute physiology and chronic health evaluation II (APACHE II) score [19.10 (8.00) vs. 20.28 (8.00), P < 0.05]. It was shown on the clinical data analysis of the hospitals that the patients meeting Sepsis-2 and Sepsis-3 in hospital 3 had the highest 28-day mortality (60.4% and 60.0%) with the lowest rate of diagnosis (0). While in the hospital 1, the patients had the lowest 28-day mortality (22.9% and 27.2%), and the rate of diagnosis was 19.5%. Interestingly, the patients in hospital 4 had the highest diagnosis rate of sepsis (44.8%), but 28-day mortality was both 58.6%. It was shown by logistic regression analysis that the patients with old age [odds ratio (OR) = 0.970, P = 0.021], high blood lactate (OR = 0.443, P = 0.004), high blood pressure (OR = 0.957, P = 0.009) and low oxygenation index (OR = 1.004, P = 0.024) were easy to neglect diagnosis. Compared with Sepsis-3 exclude group, the patients in Sepsis-3 enroll group were older [years: 68.52 (26.00) vs. 53.75 (18.00), P < 0.01] with higher APACHEII score [20.38 (8.00) vs. 7.72 (6.00), P < 0.01], higher blood lactate [mmol/L: 3.45 (3.00) vs. 1.95 (1.20), P > 0.05], longer length of ICU stay [days: 22.42 (22.00) vs. 15.13 (16.00), P < 0.01], and higher 28-day mortality [45.29% (149/329) vs. 14.71% (5/34), P < 0.01], indicating that the diagnostic efficiency of Sepsis-2 was low, the diagnostic specificity of Sepsis-3 was high, and the prognosis of Sepsis-3 patients was worse. It was shown by ROC curve analysis that the prognostic value of SIRS, qSOFA and SOFA to mortality was gradually increased [area under ROC curve (AUC) was 0.567, 0.597, 0.683, respectively], but the prognostic value were all low. Comparing patients meeting qSOFA and (or) SOFA in Sepsis-2, significant differences were found in APACHE II score [17.55 (7.00) vs. 23.24 (8.00)] and 28-day mortality [38.75% (31/80) vs. 58.59% (75/128), both P < 0.01]. The patients who just met the qSOFA or SOFA, their 28-day mortality was up to 38.75%, suggesting that qSOFA should not be ignored. Compared with survived group, the patients in survived group were older with higher APACHE II score and shorter length of ICU stay (all P < 0.05). It was shown by logistic regression analysis that APACHE II score (OR = 1.199, P = 0.000) and length of ICU stay (OR = 0.949, P = 0.000) were related with death. Patients satisfied Sepsis-3 were easier to develop more organ failure, Sepsis-3 and higher death prediction than Sepsis-2 and higher diagnosis specificity, but data shows that there is extra room for improvement.