This study was conducted to assess the prevalence of eosinophilia in 358 consecutive samples of pleural fluid (all cases corresponded to first thoracentesis), to review the cause of eosinophilic pleural effusions, and to determine whether the presence of eosinophils increases the likelihood of nonmalignant underlying disorders. Eosinophilic pleural effusions were identified in 45 patients (12.6%): malignant underlying conditions were diagnosed in 11 patients (24.4% with eosinophilic effusions) and benign aetiologies were found in 27 patients. Benign aetiologies included uncomplicated paraneumonic effusion in 10 patients, tuberculosis in seven, complicated paraneumonic in five, liver cirrhosis in three, hydronephrosis in one and pulmonary thromboembolism in one. Seven pleural effusions were idiopathic. There was no difference in the prevalence between eosinophilic and noneosinophilic effusions according to the different diagnoses. With parameters of sensitivity, specificity, pretest and post-test probability and positive and negative predictive values for any prevalence figure using the Bayes' theorem and for any value of eosinophils (both in percentage or absolute numbers) in the pleural fluid (receiver operating characteristic curve) an adequate predictor of benign disease was not found. It is concluded that pleural eosinophilia at the initial thoracentesis cannot be considered as a predictor of an underlying benign disorder.