Abstract: A clustering analysis method is used in the optimization of observed gravity data during the computation of local geoid.Two slopes of terrain data were used as criteria in order to find which data is more important according to the characteristics of gravity field.A numerical experiment was carried out in hill area.After half observed gravity data are deleted,the maximum variation of the geoid is 1.2 cm,the minimum is-0.4 cm,and the average variation is 0.3 cm.Compared with the result from non-deleted observed gravity data,the deleted data still works out acceptable results of similar precision.The numerical experiment validates the feasibility of the clustering analysis methods,which provides a new approach to optimize observed gravity data for computing local geoid.