Deconvolution has been used to correct first-pass radionuclide angiocardiography for the time course of the delivery of radiopharmaceutical into the cardiopulmonary system. The extreme sensitivity of deconvolution to random errors in the data may account for some of the problems encountered in practice. We implemented several deconvolution algorithms that were suitable for use with the unimodal and multimodal superior vena caval and pulmonary curves found in left-to-right shunt quantification. The sensitivity of the algorithms to random errors was assessed using mathematical test problems degraded with pseudorandom noise. An algorithm that constrained the deconvolved pulmonary curve to be expressable as the non-negative sum of a set of lagged normal curves was found to have the smallest maximum error on the curves tested. Comparison with results from a previously published test problem indicated an error reduction of greater than 50% over previously used algorithms. Use of this algorithm may permit more accurate deconvolution of pulmonary time-activity curves and thereby improve shunt quantification.