Purpose Diffusion tensor imaging is very sensitive to applied gradient diffusion. Because apparent diffusion coefficient (ADC) depends on the square of the amplitude of the diffusion sensitizing gradients, errors in the gradient calibration are exaggerated. The aim of this work is to propose a protocol to correct for these errors. Methods Custom spherical water phantom (14 cm diameter, filled with doped aqueous solution) was acquired. Phantom was placed at isocentre of scanner. 5 mm thick axial slices at the isocentre were acquired, using six DWI sequences (b-value 0 s/mm2 and 1000 s/mm2), differing for the gradient diffusion direction (Gdd): ±x, ±y, ±z. For each Gdd, analysis on ADC maps was performed by a custom Matlab software code. A 2× 2 cm2 ROI was considered at phantom centre and average ADC was calculated for each Gdd. Six scaling factors (one for each Gdd) were calculated as square root of ratio between reference and fitted ADCs values. A second set of three acquisitions consisting of 12, 20 and 30 gradient diffusion directions sequences were acquired: ADC and fractional anisotropy (FA) were calculated. The present protocol was repeated weekly. Results Scaling factors are between −0.9% and 0.7% and are almost constant week by week. After a periodic maintenance a significant calibration variation have been reported. Significant differences between positive and negative directions have been reported (between −0.5% and 0.4%). FA and standard deviation on measurement of the ADC from a second series of acquisitions were calculated with the corrections of the gradients, which resulted in a reduction of a factor of about two. Conclusions It was found that the six gradient coils were calibrated differently, resulting an offset on FA. The proposed protocol allows correcting for this effect. Application of this protocol to anisotropic phantom will allow better quantifying the effect of proposed corrections. Diffusion tensor imaging is very sensitive to applied gradient diffusion. Because apparent diffusion coefficient (ADC) depends on the square of the amplitude of the diffusion sensitizing gradients, errors in the gradient calibration are exaggerated. The aim of this work is to propose a protocol to correct for these errors. Custom spherical water phantom (14 cm diameter, filled with doped aqueous solution) was acquired. Phantom was placed at isocentre of scanner. 5 mm thick axial slices at the isocentre were acquired, using six DWI sequences (b-value 0 s/mm2 and 1000 s/mm2), differing for the gradient diffusion direction (Gdd): ±x, ±y, ±z. For each Gdd, analysis on ADC maps was performed by a custom Matlab software code. A 2× 2 cm2 ROI was considered at phantom centre and average ADC was calculated for each Gdd. Six scaling factors (one for each Gdd) were calculated as square root of ratio between reference and fitted ADCs values. A second set of three acquisitions consisting of 12, 20 and 30 gradient diffusion directions sequences were acquired: ADC and fractional anisotropy (FA) were calculated. The present protocol was repeated weekly. Scaling factors are between −0.9% and 0.7% and are almost constant week by week. After a periodic maintenance a significant calibration variation have been reported. Significant differences between positive and negative directions have been reported (between −0.5% and 0.4%). FA and standard deviation on measurement of the ADC from a second series of acquisitions were calculated with the corrections of the gradients, which resulted in a reduction of a factor of about two. It was found that the six gradient coils were calibrated differently, resulting an offset on FA. The proposed protocol allows correcting for this effect. Application of this protocol to anisotropic phantom will allow better quantifying the effect of proposed corrections.