Abstract: As demonstrated by an example of leveling network, existing methods of correcting gross errors based on residuals have a few shortcomings including the dependence on the location of gross errors within the leveling network. To address these problems, Local Analysis Method (LAM) is proposed. In multidimensional adjustment problem, an observed quantity usually can be expressed as function of other observed quantities. Then we can obtain a group of independent observed values of the observed quantity, which includes observed value corresponding to the observed quantity and functions of other observed values. And an approach to search functions based on the design matrix is given. So the analysis conclusion of one-dimensional problem can be used to test if the observed quantity can tolerate gross error. A method based on Combined Observed Value (function of other observed values) is also developed to detect gross errors using true errors. Finally, this detection method and LAM are detailed in an example of a goniometric network.