Lung function (LF) tests are part of many investigations in childhood lung disease. However, individual reproducibility of LF will confound between-subject differences. At the same time, increased LF variability has been linked to respiratory disease. In a sample of 598 children, two LF tests, separated by a 5-min interval, were recorded, and reliability (Rel) of forced expiratory volume in 1 s (FEV1), forced vital capacity (FVC), and maximal expiratory flow at 50% of FVC (MEF50) was determined. Rel was also assessed in children trained and untrained in the performance of LF. To investigate determinants of reproducibility for FEV1, the absolute difference between two repeated tests was calculated. Whenever this difference was > 120 ml, a child was considered to demonstrate excessive variability (poor reproducibility) in FEV1. For volume parameters coefficients of reliability (Crel) were found to be better than for MEF50 (FEV1: 0.96; FVC: 0.94, MEF50: 0.91). In untrained children Crel for FEV1 was only 0.91, but it was increased in subsequent visits (0.98, 0.97, and 0.97 at the second, third, and fourth tests, respectively). Excessive variability in FEV1 was observed in 10% of children and was related to the presence of wheeze [odds ratio (OR) 6.31; 95% confidence interval (CI) 1.78-22.4), shortness of breath (OR 3.14; 95% CI 1.00-9.93), a diagnosis of asthma (OR 6.25; 95% CI 1.76-22.1), and bronchial hyperresponsiveness (OR 4.30; 95% CI 2.07-8.94). We conclude that increased variability of LF is likely to be present in young children not accustomed to the testing procedure and in children with respiratory symptoms. Therefore, before guidelines for LF testing are applied, children should be trained to perform the tests and we should be cautious in the interpretation of test results in children who present with symptoms.