BACKGROUND: Thrombocytopenia has been shown to be a useful predictor of mortality in adult intensive care units (ICUs). The aim of this study is to assess whether the level of platelet count at ICU admission and the changes in platelet counts can predict mortality in the pediatric ICU (PICU). METHODS: Platelet counts were checked daily for at least 4 days in a total of 303 children who were admitted to the ICU. We compared the initial platelet counts and changes in platelet counts between survivors and non-survivors. A multivariable logistic regression model, a receiver operating characteristic curve and a linear mixed model were used. RESULTS: The initial platelet count was significantly lower in non-survivors when compared to survivors. Multivariate analysis demonstrated that platelet count <120 x 10(9)/L (Odds ratio, 4.913; 95% confidence interval 2.451-9.851; p < 0.0001) was an independent predictor of mortality. In the case of children with thrombocytopenia (<120 x 10(9)/L) at admission to the ICU, the platelet counts increased serially in survivors, whereas non-survivors maintained their decreased platelet counts. In the case of children without thrombocytopenia, the platelet counts decreased most on day 3 in non-survivors. CONCLUSIONS: At admission to the ICU, thrombocytopenia defined as a platelet count <120 x 10(9)/L can be a useful predictor of mortality in children. In children who had initial thrombocytopenia, the serial increase of platelet counts can be related to increased survival, whereas in children who did not have initial thrombocytopenia, more than a 10% decrease of platelet counts on day 3 can be related to mortality.