Although colloid was a component of the original Parkland formula, it has been omitted from standard Parkland resuscitation for over 30 years. However, some burn centers use colloid as "rescue" therapy for patients who exhibit progressively increasing crystalloid requirements, a phenomenon termed "fluid creep." We reviewed our experience with this procedure. With Institutional Review Board approval, we reviewed all adult patients with > or =20%TBSA burns admitted from January 1, 2005, through December 31, 2007, who completed formal resuscitation. Patients were resuscitated using the Parkland formula, adjusted to maintain urine output of 30 to 50 ml/hr. Patients who required greater amounts of fluid than expected were given a combination of 5% albumin and lactated Ringer's until fluid requirements normalized. Results were expressed as an hourly ratio (I/O ratio) of fluid infusion (ml/kg/%TBSA/hr) to urine output (ml/kg/hr). Predicted values for this ratio vary for individual patients but are usually less than 0.5 to 1.0. Fifty-two patients were reviewed, of whom 26 completed resuscitation using crystalloid alone, and the remaining 26 required albumin supplementation (AR). The groups were comparable in age, gender, weight, mortality, and time between injury and admission. AR patients had larger total and full-thickness burns and more inhalation injuries. Patients managed with crystalloid alone maintained mean resuscitation ratios from 0.13 to 0.40, whereas AR patients demonstrated progressively increasing ratios to a maximum mean of 1.97, until albumin was started. Administration of albumin produced a dramatic and precipitous return of ratios to within predicted ranges throughout the remainder of resuscitation. No patient developed abdominal compartment syndrome. Measuring hourly I/O ratios is an effective means of expressing and tracking fluid requirements. The addition of colloid to Parkland resuscitation rapidly reduces hourly fluid requirements, restores normal resuscitation ratios, and ameliorates fluid creep. This practice can be applied selectively as needed using predetermined algorithms.