Filtering by transform windows and convolution are compared. Windows are procedurally easier to design and use, especially for more complicated applications such as band-pass filtering. For comparable frequency response functions, the same amount of data will be lost at the ends of the time series regardless of which method is selected. In the case of windowing, the impulse response function must be calculated by inverting the frequency response function to determine the data loss. If large numbers of long time series are involved, computational efficiency of the filtering method becomes important. A simple index of relative efficiency is derived which shows that windowing is often but not always the more efficient technique. Even for eliminating deterministic signals such as tides, windows can sometimes improve computational efficiency when used in combination with least squares weights design.