The purpose of this study was to determine whether steady-state free precession (SSFP) could improve accuracy of geometric models for evaluation of left ventricular (LV) function in comparison to turbo gradient echo (TGrE) and thereby reduce the acquisition and post-processing times, which are commonly long by use of the Simpson's Rule. In 25 subjects, cine loops of the complete heart in short and horizontal long-axis planes were acquired using TGrE (TR/TE/flip = 5.0/1.9/25) compared with SSFP (TR/TE/flip = 3.2/1.2/60). LV volumes and EF were measured with various geometric models for TGrE and SSFP. With three-dimensional data, the LV volumes were higher and the resulting EF lower for SSFP in contrast to TGrE (51 +/- 15% vs. 57 +/- 15%, p < 0.001). With SSFP, various geometric models yielded good to excellent correlations for LV volumes and LVEF compared to volumetric data (r = 0.94-0.98, mean relative difference 7.0-11.4%). In contrast, correlations were low using biplane or single-plane ellipsoid models in TGrE (r = 0.71-0.75, mean relative difference 15.9-30.2%). A new combined geometric model, taking all three dimensions into account, yielded the highest accuracy for SSFP in comparison to volumetric data (r = 0.99, mean relative difference 4.7%). Geometric models for assessment of LV volumes and EF yield higher accuracy and reproducibility by use of the SSFP sequence than by standard TGrE. This may increase clinical utility of magnetic resonance by shorter acquisition and processing times.