Data from automatic recording systems often require editing and filtering before they are suitable for computer analysis. The procedure in this paper produces edited values at regular intervals from input data containing random noise, data gaps, and sudden steps or resets. It uses a Kalman filter with a fixed delay time to estimate the most probable data value at any time, based on information both before and after the time point. Isolated portions of a bad record can be recognized and removed, and steps or offsets are identified and measured. An example is shown of clean output produced from input which suffers from a variety of instrumental problems.--Modified journal abstract.