BACKGROUND: To devise an improved blood-oxygen-level-dependent (BOLD) imaging protocol for cerebrovascular reactivity (CVR) measurement that can remove a known artifact of negative values. METHODS: Theoretical and simulation studies were first performed to understand the biophysical mechanism of the negative CVR signals, through which improved BOLD sequence parameters were proposed. This was achieved by equating signal intensities between cerebrospinal fluid and blood, by means of shortening the echo time (TE) of the BOLD sequence. Then, 10 healthy volunteers were recruited to participate in an experimental study, in which we compared the CVR results of two versions of the optimized ("Opt1" and "Opt2") protocols with that of the standard protocol at 3 Tesla. Two sessions were performed for each subject to test the reproducibility of all three protocols. RESULTS: Experimental results demonstrated that the optimized protocols resulted in elimination of negative-CVR voxels. Quantitative CVR results were compared across protocols, which show that the optimized protocols yielded smaller CVR values (Opt1: 0.16 ± 0.01 %BOLD/mmHg CO2 ; Opt2: 0.15 ± 0.01 %BOLD/mmHg CO2 ) than (P < 0.001) the standard protocol (0.21 ± 0.01 %BOLD/mmHg CO2 ), but the CNR was comparable (P = 0.1) to the standard protocol. The coefficient-of-variation between repetitions was found to be 5.6 ± 1.4%, 6.3 ± 1.6%, and 6.9 ± 0.9% for the three protocols, but there were no significant differences (P = 0.65). CONCLUSION: Based on the theoretical and experimental results obtained from this study, we suggest that the use of a TE shorter than those used in fMRI is necessary to minimize negative artifact in CVR results.