Whether conventional gradient-echo (GE) blood oxygenation-level-dependent (BOLD) functional magnetic resonance imaging (fMRI) is able to map submillimeter-scale functional columns remains debatable mainly because of the spatially nonspecific large vessel contribution, poor sensitivity and reproducibility, and lack of independent evaluation. Furthermore, if the results from optical imaging of intrinsic signals are directly applicable, regions with the highest BOLD signals may indicate neurally inactive domains rather than active columns when multiple columns are activated. To examine these issues, we performed BOLD fMRI at a magnetic field of 9.4 tesla to map orientation-selective columns of isoflurane-anesthetized cats. We could not convincingly map orientation columns using conventional block-design stimulation and differential analysis method because of large fluctuations of signals. However, we successfully obtained GE BOLD iso-orientation maps with high reproducibility (r = 0.74) using temporally encoded continuous cyclic orientation stimulation with Fourier data analysis, which reduces orientation-nonselective signals such as draining artifacts and is less sensitive to signal fluctuations. We further reduced large vessel contribution using the improved spin-echo (SE) BOLD method but with overall decreased sensitivity. Both GE and SE BOLD iso-orientation maps excluding large pial vascular regions were significantly correlated to maps with a known neural interpretation, which were obtained in contrast agent-aided cerebral blood volume fMRI and total hemoglobin-based optical imaging of intrinsic signals at a hemoglobin iso-sbestic point (570 nm). These results suggest that, unlike the expectation from deoxyhemoglobin-based optical imaging studies, the highest BOLD signals are localized to the sites of increased neural activity when column-nonselective signals are suppressed.