We investigated the role of dynamic changes of serum levels S100B protein in brain injury and poor outcome of sepsis. This is a prospective cohort study designed to include 104 adult patients with sepsis who are admitted to ICU from Jan 2015 to Aug 2016. Sepsis was defined as sepsis 3.0. Patients with a GCS score of <15, or at least one positive CAM-ICU score were thought to have brain dysfunction. 59 patients were diagnosed with SAE and the rest 45 patients were diagnosed with non-SAE. Serum S100B was measured on day 1 and 3 after ICU admission. Primary outcomes included brain dysfunction and 28-day/180-day mortality. The SAE group showed a significantly higher APACHE II score, SOFA scores, length of ICU stay, 28-day and 180-day mortality, serum S100B levels on day 1 and day 3. S100B levels on day 1 of 0.226 μg/L were diagnostic for SAE with 80.0% specificity and 66.1% sensitivity, and the area under (AUC) the curve was 0.728, S100B levels on day 3 of 0.144 μg/L were diagnostic for SAE with 84.44% specificity and 69.49% sensitivity, and the AUC was 0.819. In addition, the AUC for S100B on day 3 for predicting 180-day mortality was larger than for S100B on day 1 (0.731 vs. 0.611). Multiple logistic regression analysis showed that S100B3 (p = 0.001) but not S100B1 (p = 0.927) were independently correlated with SAE. Kaplan-Meier survival analysis showed that patients with S100B levels higher than 0.144 μg/L had a lower probability of survival at day 180. There were more patients with encephalopathy and a higher 28-day or 180-day mortality in the ΔS100B + group than in the ΔS100B- group. Multiple logistic regression analysis showed that SAE and IL-6 on day 3 were independently correlated with S100B dynamic increase. These findings suggest that elevated serum S100B levels on day 3 and the dynamic changes of serum S100B levels from day three to one were more associated with brain dysfunction and mortality than that on day 1 in patients with sepsis.