Inductance cardiography (thoracocardiography) non-invasively monitors changes in stroke volume by recording ventricular volume curves with an inductive plethysmographic transducer encircling the chest at the level of the heart. Clinical application of this method has been hampered, as data analysis has not been feasible in real time. Therefore a novel, real-time signal processing technique for inductance cardiography has been developed. Its essential concept consists in performance of multiple tasks by several, logically linked signal processing modules that have access to common databases. Based on these principles, a software application was designed that performs acquisition, display, filtering and ECG-triggered ensemble averaging of inductance signals and separates cardiogenic waveforms from noise related to respiration and other sources. The resulting ventricular volume curves are automatically analysed. Performance of the technique for monitoring cardiac output in real time was compared with thermodilution in four patients in an intensive care unit. The bias (mean difference) among 76 paired thoracocardiographic and thermodilution derived changes in cardiac output was 0%; limits of agreement (+/- 2 SD of the bias) were +/- 25%. It is concluded that the proposed signal processing technique for inductance cardiography holds promise for non-invasive, real-time estimation of changes in cardiac output.