The ability to predict the outcome in acute tubular necrosis (ATN) remains elusive despite considerable efforts. Accurate prediction is a crucial priority and has large economical and ethical implications, mainly to judge when treatment is futile and further efforts only prolong miserable agony. To analyze the influence of risk factors in the prognosis of ATN, we applied, in an initial phase, a prospective protocol of demographic data, cause of renal failure, diuresis, need of dialysis and clinical conditions in 228 patients using multiple linear and logistic regression models. In a control phase with 100 consecutive patients, we checked the accuracy of the results previously obtained, evaluating further the overall population of 328 patients in a synthetic phase. Finally, the validation of the equations obtained was verified in 25 patients from another hospital. As a complement of this 4-phase study, detailed statistical comparisons between both linear and logistic multiple regression models were undertaken. Correlation between probability of death obtained with equations from the initial phase applied to control patients and real evolution of these patients, survival or death, was excellent. The study of the synthetic phase revealed coma, assisted respiration, hypotension, oliguria and jaundice as having an independent positive influence on mortality and nephrotoxic etiology and normal consciousness on good prognosis. For the linear model, the same cut-off point of discriminant score (0.9) above which there were no chances for survival could be established in the 4 phases. With the logistic model, it only was found at later phases. The multiple linear was better than the logistic regression model in terms of better correlation with real mortality, better sensitivity and specificity intervals, easier use of discriminant cut-off point and better adjustment of distribution of standardized residuals to expected normal function. Early prognosis of ATN is possible and can be given using simple clinical features. A discriminant score allows to distinguish patients without chances for survival. The multiple linear is better than the logistic regression model in the prediction of the outcome in ATN.