This study investigates the use of real-time acquisition in cardiac magnetic resonance imaging (MRI) for measurements of left ventricular dimensions in comparison with conventional gradient echo acquisition. Thirty-one subjects with a variety of left ventricular morphologies to represent a typical clinical population were studied. Short-axis data sets of the left ventricle (LV) were acquired using a conventional turbo-gradient echo and an ultrafast hybrid gradient echo/echo planar sequence with acquisition in real-time. End-diastolic volume (EDV), end-systolic volume (ESV), ejection fraction (EF) and left ventricular mass (LV mass) were measured. The agreement between the two acquisitions and interobserver, intraobserver and interstudy variabilities were determined. The bias between the two methods was 5.86 ml for EDV, 0.23 ml for ESV and 0.94% for EF. LV mass measurements were significantly lower with the real-time method (mean bias 14.38 g). This is likely to be the result of lower spatial resolution and chemical shift artefacts with the real-time method. Interobserver, intraobserver and interstudy variabilities were low for all parameters. In conclusion, real time acquisition in MRI can provide accurate and reproducible measurements of LV dimensions in subjects with normal as well as abnormal LV morphologies, but LV mass measurements were lower than with conventional gradient echo imaging.