Recent research in several fields such as Biotechnology and Healthcare has uncovered a vast number of applications where 3D Electron Microscopy (EM) is useful. However, images produced by 3D EM are in most cases severely degraded. These degradations arise due to a multitude of reasons, e.g. the complex electronics in the system, magnetic lens aberration, heating and motion stability, charging, etc. Although the raw, degraded images are currently used for analysis, their usefulness is limited because the degradations make visual distinction and automated analysis of biological features very difficult. In this work, we give an analysis of noise, as one of the most important degradations in 3D EM imaging. Next, we propose a Non-Local Means image restoration algorithm that exploits the derived noise characteristics. The proposed algorithm yields significant improvements compared to other state-of-the-art image restoration algorithms.