At each node of a network-type survey, two surveyed values are obtained. In the absence of errors, these two values should be equal, providing that the quantity being surveyed is, for all practical purposes, time independent. When a low-frequency, time-dependent error enters the system, it can be indirectly observed in the form of surveyed-value differences occurring at the network nodes. This paper describes a technique for reducing such an error. The method consists of using a conjugate gradient projection algorithm to find the unbiased, discrete function of minimum weighted variation which produces the observed nodal differences. After inspecting the discrete solution, a suitably chosen continuous function can be fitted to the discrete data and used to reduce the time-dependent error. This technique has been successfully used to reduce the radial ephemeris error present in SEASAT geoid height data (Cloutier, 1981). Future applications include the removal of the diurnal variation present in magnetic surveys and the removal of the nonlinear gravimeter drift present in gravity surveys.