Shows that the rules of thumb for propagating significant figures through arithmetic calculations frequently yield misleading results. Also describes two procedures for performing this propagation more reliably than the rules of thumb. However, both require considerably more calculational effort than do the rules. (JN)