Fraction arithmetic facility is fundamental to learning more advanced math topics. However, attaining the ability to add and subtract fractions is hard for many students. The present longitudinal study examined students' growth on simple addition and subtraction word problems between fourth and sixth grades (N = 536). Latent class growth analyses revealed three empirically distinct growth trajectories: (a) "consistently accurate" students who displayed high accuracy across time; (b) "high-growth" students who demonstrated low initial accuracy but made subsequent growth; and (c) "low-growth" students who demonstrated consistently poor accuracy across time. Age and whole number calculation fluency predicted membership in the consistently accurate and high-growth classes relative to membership in the low-growth class. Language and nonverbal reasoning skills were identified as predictors of student membership in the consistently accurate versus low-growth class. Low-growth students displayed more whole number bias and calculation errors compared with students who demonstrated growth. Findings reveal that a concerning subset of students make little to no progress in simple fraction arithmetic computation, despite several years of fractions instruction in school.