Coordinate measuring machine (CMM) plays a more and more important role in the flatness evaluation. The pre-processing for eliminating error data is an important process to improve the quality of flatness evaluation. So, flatness uncertainty estimation becomes a key problem. To solve the problem, a flatness uncertainty estimation method based on data elimination is presented. Firstly, a method for excluding the error data based on statistical theory is expatiated with the consideration of probability theory and mathematical statistics. Secondly, details of calculating flatness uncertainty based on least-square method are analyzed. Then the measured 3-dimensional data are processed so as to put the excluding method into use. So the error data can be deleted and the flatness uncertainty can be decreased and can achieve a good evaluation quality. Finally, an example is given and the result shows that the method proposed in this paper is feasible.