Reproducibility of thermal-dye extravascular lung water (EVLW) estimation by a lung water computer was studied by performing ten consecutive measurements in 45 critically ill patients. EVLW ranged over a wide spectrum from 187 to 1163 ml (2.4 to 18.6 ml/kg). The mean coefficient of variation of ten consecutive measurements was 13.6%. Spontaneously breathing patients showed significantly (p less than .05) higher coefficients of variation (16.1 +/- 6.8%) than patients on mechanical ventilation (10.8 +/- 4.2%). Other factors affecting reproducibility could not be clearly identified. Because in lung water estimation mean values of consecutive measurements are compared, we defined an EVLW determination as the mean value of three consecutive EVLW measurements. When comparing consecutive EVLW determinations in hemodynamically stable patients, we found in many that consecutive EVLW determinations varied no more than +/- 15%. These differences probably have to be attributed to the reproducibility of EVLW measurements and have to be considered, when changes in thermal-dye lung water measured by a lung water computer are discussed.