We have investigated the accuracy and reproducibility of left ventricular (LV) functions using quantitative, gated, single photon emission computed tomography (SPECT) software dependent on critical frequencies of pre-reconstruction filters. This study incorporated dynamic myocardial phantoms (myocardial and cone shapes). Gated SPECT of 8-interval sets were pre-filtered with Butterworth filters (critical frequency varying between 0.16 and 1.16 cycles/cm, order 5) and with no filter. Phantoms were repositioned, and SPECT acquisitions were repeated. As the critical frequency increased, the estimated LV volume increased to reach a plateau at the level of the critical frequency, 0.54 cycles/cm. Conversely, the values of ejection fractions, wall motion and wall thickening with different filters which used critical frequency of > or =0.39 cycles/cm remained unchanged. However, LV functions and volumes were underestimated when any pre-reconstruction filter was used. Standard deviations of LV functions after repeated measurements were unaffected by different filters with critical frequencies of > or =0.39 cycles/cm. Standard deviations of LV volume, ejection fraction, wall motion and wall thickening were <2.2 ml, <0.9%, <0.6 mm and <8.7%, respectively. Therefore, with the exception of low critical frequencies, LV functions and volumes were highly reproducible when these routine reconstruction filters were used.