A system was developed for objective prediction of survival, which could be applied to all critically ill patients, regardless of disease, at an early stage of hospitaluation. Such a system would allow risk assessment of groups for experimental studies according to probability of survival and it might allow us to avoid expenditure of scarce clinical resources on animals with little chance of survival. The prospective study included 200 critically ill dogs: 93 females and 107 males, representing 62 breeds. With survival defined as alive 30 days after admisssion to ICU the overall mortality rate was 40.5% (81 of 200 dogs). Data collected included signalment and parameters that reflected vital organ function, the severity of physiologic derangement and the extent of physiologic reserve. We recorded the most abnormal value for each parameter within 24 hours of admission to ICU and logistic regression analysis was used to analyze four different weighting systems. The best model had a concordance 0f 86.5% with outcome, and it was then re-evaluated to determine whether individual variables could be eliminated without losing predictive accuracy. Fourvariables were eliminated, resulting in a final model with 18 variables which had 86.3% concordance with outcome. At a 0.5 cut-off for predicted risk, the model had sensitivity Of 69%, specificity ficity of 86% and positive predictive value of 77%. A Receiver Operating Curve was constructed using serial cut-offs for predicted outcome from 0.1 to 0.9, and the area under the curve was 0.89. Thus, an equation was generated that gives an estimate of the probability of survival.