objective: the aim of this study is to evaluate the applicability of guaragnascore for predicting mortality in patients undergoing heart valve surgery in the division of cardiovascular surgery of pronto socorro cardiológico de pernambuco - procape, recife, pe, brazil. methods: retrospective study involving 491 consecutive patients operated between may/2007 and december/2010. the registers contained all the information used to calculate the score. the outcome of interest was death. association of model factors with death (univariate analysis and multivariate logistic regression analysis), association of risk score classes with death and accuracy of the model by the area under the roc (receiver operating characteristic) curve were calculated. results: the incidence of death was 15.1%. the nine variables of the score were predictive of perioperative death in both univariate and multivariate analysis. we observed that the higher the risk class of the patient (low, medium, high, very high, extremely high), the greater is the incidence of postoperative af (0%; 7.2%; 25.5%; 38.5%; 52.4%), showing that the model seems to be a good predictor of risk of postoperative death, in a statistically significant association (p <0.001). the score presented a good accuracy, since the discrimination power of the model in this study according to the roc curve was 78.1%. conclusions: the brazilian score proved to be a simple and objective index, revealing a satisfactory predictor of perioperative mortality in patients undergoing heart valve surgery at our institution.