Phase-contrast MRI (PC-MRI) of cerebrospinal fluid (CSF) velocity is used to evaluate the characteristics of intracranial diseases, such as normal-pressure hydrocephalus (NPH). Nevertheless, PC-MRI has several potential error sources, with eddy-current-based phase offset error being non-negligible in CSF measurement. In this study, we assess the measurement error of CSF velocity maps obtained using 4D flow MRI and evaluate correction methods. CSF velocity maps of 10 patients with NPH were acquired using 4D flow MRI (velocity-encoding = 5 cm/s). Distributed phase offset error was estimated for a whole 3D background field by polynomial fitting using robust regression analysis. This estimated phase offset error was then used to correct the CSF velocity maps. The estimated error profiles were compared with those obtained using an existing 2D correction approach involving local background information near the region of interest. The residual standard error of the polynomial fitting against the phase offset error extracted from the measured velocities was within 0.2 cm/s. The spatial dependencies of the phase offset errors showed similar tendencies in all cases, but sufficient differences in these values were found to indicate requirement of velocity correction. Differences of the estimated errors among other correction approaches were in the order of 10-2 cm/s, and the estimated errors were in good agreement with those obtained using existing approaches. Our method is capable of estimating the measurement error of CSF velocity maps obtained from 4D flow MRI and provides quantitatively reasonable characteristics for the main CSF profile in the cerebral aqueduct in patients with NPH.