This study is a retrospective survey of the variables that may influence the development of pneumothorax after thoracentesis. In a 30-month period, a computer search of hospital records identified 342 thoracenteses, of which 154 were done with conventional techniques by the clinical services, and 188 were done with sonographic guidance. Other factors surveyed included the patients' age, sex, underlying pulmonary disease, and overall clinical condition; the size of the effusion; the type of tap (diagnostic or therapeutic); the amount and type (exudate or transudate) of fluid acquired; and the size of the needles used. The technique used was the most significant single risk factor affecting the development of pneumothorax (18% for clinical vs 3% for sonography-guided thoracenteses). The incidence of pneumothorax decreased when a smaller amount of pleural fluid was aspirated (mean, 246 ml aspirated from patients who did not vs 472 ml from those who did develop pneumothorax) and when thin needles were used (4% pneumothorax with 20-gauge or smaller and 18% with larger than 20-gauge needles). The other factors surveyed did not influence the development of pneumothorax. Our results show that sonography-guided thoracentesis is complicated by pneumothorax significantly less often than is thoracentesis done with conventional techniques. Use of the smallest possible needle and aspiration of the smallest possible amount of fluid will also result in fewer cases of pneumothorax.