AIMS: Heart failure (HF) patients are at high-risk of cardiovascular (CV) events, including CV death. Nonetheless, a substantial proportion of these patients die from non-CV causes. Identifying patients at higher risk for each individual event may help selecting patients for clinical trials and tailoring cardiovascular therapies. The aims of the present study are to: (i) characterize patients according to CV vs. non-CV death; (ii) develop models for the prediction of the respective events; (iii) assess the models' performance to differentiate CV from non-CV death. METHODS AND RESULTS: This study included 2309 patients with HF from the BIOSTAT-CHF (a systems BIOlogy Study to TAilored Treatment in Chronic Heart Failure) study. Competing-risk models were used to assess the best combination of variables associated with each cause-specific death. Results were validated in an independent cohort of 1738 HF patients. The best model to predict CV death included low blood pressure, estimated glomerular filtration rate ≤ 60 mL/min, peripheral oedema, previous HF hospitalization, ischaemic HF, chronic obstructive pulmonary disease, elevated N-terminal pro-B-type natriuretic peptide (NT-proBNP), and troponin (c-index = 0.73). The non-CV death model incorporated age > 75 years, anaemia and elevated NT-proBNP (c-index = 0.71). Both CV and non-CV death rose by quintiles of the risk scores; yet these models allowed the identification of patients in whom absolute CV death rates clearly outweigh non-CV death ones. These findings were externally replicated, but performed worse in a less severely diseased population. CONCLUSIONS: Risk models for predicting CV and non-CV death allowed the identification of patients at higher absolute risk of dying from CV causes (vs. non-CV ones). Troponin helped in predicting CV death only, whereas NT-proBNP helped in the prediction of both CV and non-CV death. These findings can be useful both for tailoring therapies and for patient selection in HF trials in order to attain CV event enrichment.