The ability of chest radiographs to determine the size of a pneumothorax was tested in 16 patients using computed tomographic (CT) scan as a reference method. To determine if CT with a slice thickness of 12 mm could be used, its accuracy was assessed in a lung model experiment. The lung model consisted of a water-filled plastic bag (lung) fitted into a plastic chamber (hemithorax), both of approximately the same size and shape as in man. Water was drawn off in incremental steps and a CT was done after each step. The area of the pneumothorax was calculated by computer and when multiplied by slice thickness and number of slices, the total volume of the pneumothorax could be compared with the extracted amount of water. A good correlation (r = 0.99), with the line of regression close to the line of identity, was found between the CT investigation and the artificial pneumothorax. In the patients, the size of the pneumothorax, judged by radiograph using two different methods of calculation, was correlated to the size obtained by CT. The correlation was poor (r = 0.71) irrespective of method of calculation. The size of the pneumothorax estimated by CT showed a good correlation (r = 0.99) to the initial aspirated air volumes in 12 of the 16 patients treated with drainage. A cautious attitude toward the use of chest radiographs for calculations of the degree of lung collapse in patients with pneumothorax is recommended.