This study tests the hypothesis that a change in hematocrit (ΔHct) during initial trauma work-up is as reliable as conventional vital signs for detecting bleeding, even with ongoing fluid resuscitation. Consecutive trauma patients admitted to a Level I trauma center receiving two Hct measurements during initial resuscitation between January 2010 and January 2011 were stratified based on estimated blood loss greater than 250 mL (bleeding) or nonbleeding. Sensitivity, specificity, and receiver operating characteristic curves were calculated for systolic blood pressure (SBP), heart rate, base deficit, and ΔHct. In 168 (72%) nonbleeding versus 64 (28%) bleeding patients, age and gender were similar. Arrival SBP was highly specific (90 to 99%) but poorly sensitive (13 to 31%) for detecting bleeding. Combinations of vital signs increased specificity, albeit at the expense of sensitivity. For bleeding versus nonbleeding patients (all receiving resuscitation fluid), ΔHct was 9.0 versus 1.8, ΔHct/liter was 4.8 versus 1.5, and ΔHct/liter/hour was 2.8 vs 0.6 (all P < 0.001). Only SBP (area under the curve [AUC] 0.608 to 0.695) and ΔHct (AUC 0.731 to 0.921) were significant for identifying bleeding with ΔHct 6 or greater being the best predictor (sensitivity 89%, specificity 95%, AUC 0.921). During ongoing fluid resuscitation of a trauma victim, ΔHct is the single most reliable indicator of continuing blood loss. A ΔHct 6 or greater during initial resuscitation is highly suspicious for ongoing blood loss, but even lesser changes have predictive value. Altogether, these results support the idea that fluid shifts are rapid after hemorrhage and Hct can be valuable during initial trauma assessment.