In a prospective study of 334 consecutive patients with chronic pleural effusion 156 or 47% had a malignant etiology. The sensitivity of cytological examination of pleural fluid in detecting malignancies was 43%. Thoracoscopy had a sensitivity of 80% and could reveal malignancy in 37 of 47 patients with malignant effusions and a negative cytology. A malignant etiology must also be suspected when atypical cells are found in the pleural fluid as in our study 16 out of 19 fluid samples with atypical cytology represented malignant effusions. The predictive value for a malignant etiology was estimated for the following variables: sex, size of effusion, colour and eosinophils of pleural fluid, smoking habits and asbestos exposure. The predictive value of each variable was estimated separately, combining two by two and by a logistic regression function to exclude correlation to yet another variable. The single variable with the strongest positive predictability towards malignancy was a bloody fluid. Conversely, more than 30 per cent eosinophils in the fluid had the strongest negative predictability towards malignancy. The ability of our statistical method, a logistic regression function to discriminate between malignant and non-malignant etiology was 79%. The estimated probability of a malignant etiology should influence the choice of invasive procedures and the duration and intensity of follow-up.