Recent reductions of Apollo subsatellite and Lunar Orbiter 5 data have determined the first plausible models for the farside lunar gravity field. This paper presents a selenodesy method which estimates gravity by fitting to the long-term variations of the Kepler element rates. Raw Doppler tracking data taken over short arcs are reduced to estimate a best set of mean orbital elements for each orbit. A succession of such fits in performed to generate a history of mean elements. The element rates are determined from patched cubic spline fits to the elements. The rates are adjusted for n-body effects and along with the associated elements are used as input to a gravity estimator. This method eliminates certain of the dynamical aspects of long-term selenodesy, and consequently, the gravity inversion is a linear process. Since the rates are generated from a series of patched cubic spline fits, unmodeled spacecraft maneuvers contaminate at most only several data points, and there is no significant net integrated effect as in conventional long-term methods. Simulations performed demonstrate that farside gravity features can successfully be determined by fitting to mean elements derived from nearside tracking. Arguments are presented which conclude that a long-term gravity method of this type is the most plausible technique which can obtain realistic estimates for farside lunar gravity using the currently available data.