The use of non-invasive ventilation (NIV) in severe acute respiratory failure (ARF) due to community-acquired pneumonia (CAP) is controversial, and the risk factors for NIV failure in these patients are not well known. We assessed the characteristics and predictors of outcome of patients with CAP and severe ARF treated with NIV.We prospectively assessed 184 consecutive patients; 102 had "de novo" ARF, and 82 previous cardiac or respiratory disease. We defined successful NIV as avoidance of intubation and intensive care unit (ICU) survival at least 24 h in the ward. We assessed predictors of NIV failure and hospital mortality in multivariate analyses.Patients with "de novo" ARF failed NIV more frequently than patients with previous cardiac or respiratory disease (47, 46% versus 21, 26%, p = 0.007). Worsening radiologic infiltrate 24 h after admission, maximum Sepsis-Related Organ Failure Assessment (SOFA) score and, after 1 h of NIV, higher heart rate and lower PaO(2)/FiO(2) and bicarbonate independently predicted NIV failure. Likewise, maximum SOFA, NIV failure and older age independently predicted hospital mortality. Among intubated patients with "de novo" ARF, NIV duration was shorter in hospital survivors than non-survivors (32 ± 24 versus 78 ± 65 h, p = 0.014). In this group, longer duration of NIV before intubation was associated with decreased hospital survival (adjusted odds ratio 0.978, 95% confidence interval 0.962-0.995, p = 0.012). This association was not observed in patients with previous cardiac or respiratory disease.Successful NIV was strongly associated with better survival. If predictors for NIV failure are present, avoiding delayed intubation of patients with "de novo" ARF would potentially minimise mortality.