A large earthquake (M7.1) occurred during the 1914 eruption of Sakurajima volcano in Kagoshima prefecture, Japan. I estimated seismic intensities in the present Japan Meteorological Agency (JMA) scale for the 1914 Sakurajima earthquake in Kagoshima city. Previous studies on seismic intensities in Kagoshima city are used their own unique methods to estimated seismic intensities from data of damaged houses and stonewalls. I used a method of estimating seismic intensities from data of damaged houses, which has been commonly used. Previous studies using this method have assumed that each household owns one house, but the numbers of damaged houses in some towns are significantly larger than those of households in the case of the 1914 Sakurajima earthquake. I, therefore, made two assumptions A and B. The assumption A is that each household owns one house. The assumption B is that the ratio of houses to households in each town equal to the maximum ratio of total damaged houses to households. As a result, the maximum seismic intensities for the assumptions A and B are 6 upper and 6 lower, respectively. The areas with higher seismic intensity than 6 lower are consistent with the previous studies on the 1914 Sakurajima earthquake, but the areas with lower seismic intensities than 5 upper are inconsistent because I did not take stonewall damages into consideration, which the previous studies did. I compared our results with distributions of seismic intensities predicted by an attenuation relation between intensity and distance with two different sets of source parameters. One set of source parameters showed seismic intensities of 6 lower and 6 upper in Kagoshima city, and the other set showed those of 5 lower and 5 upper. The observed seismic intensities are on average lower than the predicted ones, probably suggesting that large coseismic slip area is distributed further than epicenter, that Kagoshima city is located in the opposite direction of the rupture propagation and its directivity effect make the amplitudes smaller, or that amplitudes at frequencies which affect seismic intensities are smaller than those expected from the magnitude. The observed seismic intensity distributions are also rougher than the predicted ones and could be affected by local soil conditions.