In patients with cystic fibrosis (CF), pulmonary exacerbations (PEx) have an important influence on well-being, quality of life, and lung function decline. Early detection combined with early treatment may prevent severe PEx. To determine whether early detection of PEx is possible by non-invasive markers (volatile organic compounds) in exhaled breath. In a 1 year prospective observational pilot study, 49 children with CF were studied. At clinical visits with an interval of 2 months, lung function, volatile organic compounds (VOCs) in exhaled breath by means of gas chromatography-time-of-flight-mass spectrometry, and medication use were assessed. PEx were recorded. Random forest (RF) classification modelling was used to select discriminatory VOCs, followed by building of receiver operating characteristic curves. An inverse relation between the predictive power of a set of VOCs and time between exhaled breath sampling and the onset of PEx was found. When this time period was within 7 d, the RF model with the nine most discriminatory VOCs was able to correctly predict 79% of the children with an upcoming PEx or remaining stable (sensitivity 79% and specificity 78%). This result was validated by means of bootstrapping within the RF classification model. PEx in children with CF can be detected at an early stage by means of exhaled VOCs. The highest predictive value was reached if time between sampling and the onset of an exacerbation was no longer than 7 d.