A unit on adding and subtracting fractions based on the formal definition from rational numbers was developed and compared to use of an unspecified method. Eighty-one fifth-grade students were randomly assigned to two classes for arithmetic instruction. After completion of the unit, an immediate and a delayed posttest were given; an analysis of variance and of covariance were used on these respectively. Findings showed that the experimental group performed significantly better on both criterion tests, but that the control group wrote significantly more correct answers in lowest terms. (Author/DT)