When the mean birth weights (MBW) calculated from frequency tables are compared with each other, the methods for grouping data into class intervals should be taken into consideration. We integrated the lower and/or upper classes of the frequency tables belonging to the type used in the Vital Statistics of Japan with 500 g intervals and 10 classes (less than or equal to 999, 1000-1499,..., 4500-4999, 5000 less than or equal to) into neighboring classes, and observed the differences between both MBW and the standard deviations (SD) calculated from the original standard frequency tables (o method) and those from the above modified ones (a,..., h method) at three community levels, namely, prefectural, health center (HC) and city-town-village level. At the prefectural level, MBW changed from -2 to 3 g through all the modified methods and SD changed most at -5 g in a,..., d method. At the HC level, MBW changed from -3 to -6 g and SD changed most at -8 g in a,..., d method. At the city-town-village level, MBW and SD did not change in over 50% of the communities, but the degrees of change were large in some communities where MBW or SD changed. The changes were larger in the case of integrating the lowest class into a class of less than or equal to 1499 g than that of integrating the highest class into a class of 4500 less than or equal to. Also they were larger in the case of integrating the lower two classes into a class of less than or equal to 1999 g than that of integrating the upper two classes into a class of 4000 g less than or equal to. We concluded that 1) MBW from the frequency tables where the lower classes were integrated should be carefully dealt with, especially at the small community level, but 2) the differences of population means between any method were not statistically significant and 3) if MBW from different frequency tables are compared with each other, reasonable differences of MBW depending on the methods of grouping and the size of the number of live births should be taken into account.