Abstract It has been reported that unusually large errors are observed in velocities derived from EISCAT data in which the power in the returned signal falls to less than 2% of that in the background noise. This suggestion has been investigated using a specialised interactive analysis package, and possible causes, both in the data and in the analysis, were considered. Subsequently, an adaptation of the usual procedure for velocity determination was considered. This involved applying the normal method to the raw data without first subtracting the background noise. The advantages of this technique are discussed, in particular for data in which the returned signal is relatively weak.