The CLEAN deconvolution algorithm has well-known limitations due to the restriction of locating point source model components on a discretized grid. In this letter we demonstrate that these limitations are even more pronounced when applying CLEAN in the case of Rotation Measure (RM) synthesis imaging. We suggest a modification that uses Maximum Likelihood estimation to adjust the CLEAN-derived sky model. We demonstrate through the use of mock one-dimensional RM synthesis observations that this technique shows significant improvement over standard CLEAN and gives results that are independent of the chosen image pixelization. We suggest using this simple modification to CLEAN in upcoming polarization sensitive sky surveys.