A new method is presented for the depth interpolation of hydrographic data below the level of direct seasonal influence. It consists of interpolating (here, linearly) the departures of individual station data from a mean profile that has been estimated by cubic spline smoothing from an ensemble of stations occupied at the same or nearby locations. Repeated temperature profiles taken at the Panulirus station off Bermuda are used for illustration. The rms interpolation error is estimated for a very large number of temperature measurements between 220 and 2000 m and compared to rms interpolation errors obtained with other methods. The new scheme performs best, particularly in data sparse regions. The interpolation method currently used at the National Oceanographic Data Center and the Aitken-Lagrange interpolation yield a 30% larger rms interpolation error; the interpolation error for linear interpolation is more than 80% larger. The applicability of the interpolation method using the mean profile is also tested in the case where only a few stations are available to define the mean. Although the error reduction is smaller, the method still performs best when sets of four or more stations are available.