OBJECTIVE To analyze the reproducibility of LV volumes calculated by cardiac magnetic resonance imaging (CMRI) and to compare them to those obtained by conventional ventriculography. METHODS A total of 30 patients with stable ischemic heart disease were prospectively included. Each underwent CMRI twice and ventriculography. Left ventricular end diastolic volume (EDV), end systolic volume (ESV) and LV ejection fraction (EF) were calculated by two radiologists at different level of experience. Intraobserver, interobserver and interstudy variabilities were assessed. RESULTS The cut off values were: intraobserver variability (EDV, ESV, EF): 9.4 ml, 5.3 ml, 3.3% for well-trained radiologist; 13.1 ml, 7.5 ml, 4.1% for less-trained radiologist. interobserver variability: EDV: 11.7 and 10.4 ml; ESV: 7.0 and 6.6 ml; EF: 3.9 and 4.2%. interstudy variability (EDV, ESV, EF): 11.6 and 12.6 ml, 7.1 and 7.4 ml, 3.9 and 3.5%, for experienced and less-trained observers. Statistical differences were found between CMRI and ventriculography: CMRI underestimation of EDV and EF, overestimation of ESV. CONCLUSIONS CMRI volumetric quantification of LV volumes and function is highly reproducible at different levels of experience, but not interchangeable with those obtained by ventriculography.