Lunar phases of earthquakes worldwide with M_W ≥ 8 between 1900 and November 15, 2016, were calculated. The null hypothesis that the occurrences of the giant earthquakes are random was assumed and tested for some subduction zones and around Mongolia. The null hypothesis was statistically rejected for Southern Chile, Alaska, Peru, the Japan Trench, the Kuril Islands, and Around Mongolia. The rejection of the random null hypothesis means that the occurrences of the giant earthquakes are not random. However, there are statistically significant specific lunar phase ranges (dangerous lunar phase, DLP) in which the giant earthquakes concentrate for each subduction zone and around Mongolia. DLPs for Southern Chile, Alaska, and Peru are almost around the spring tides. However, DLPs of the Japan Trench, the Kuril Islands, and around Mongolia are rather around the neap tides. A method to evaluate the DLP for the next giant earthquake with a significance level of 0.01 and 0.05 was proposed. The lower limit of MW for the analysis was investigated for Peruvian earthquakes, and MW = 7.5 was considered the best. The DLP was calculated for the giant earthquakes in the Japan Trench before Tohoku 2011. It was found that the foreshocks of Tohoku 2011 and the mainshock occurred in the DLP. Heightened attention during the DLP in each seismic zone, primarily when seismicity occurs, should reduce the damage from giant earthquakes.