The stability of the Gross Motor Function Classification System (GMFCS) over time is described in 77 infants (41 boys, 36 girls) with cerebral palsy (CP; mean age 19.4mo [SD 1.6 mo]; 27 unilateral spastic, 42 bilateral spastic, eight dyskinetic type) and in the same children at follow-up at age 2 to 4 years. The overall level of agreement over time (linear weighted kappa) was 0.70 (95% confidence interval [CI] 0.61-0.79). The overall percentage of children whose GMFCS level changed one or two levels was 42%, of which the majority were reclassified to a less functional level (McNemar's Chi(2) test p=0.11). The chance that children initially classified in the combination of GMFCS Levels I, II, and III would subsequently be classified in the same level in early childhood was 96% (positive predictive value [PPV] 0.96, 95% CI 0.85-0.99), whereas the PPV for the combination of Levels I and II was 0.88, 95% CI 0.70-0.96. These findings indicate that GMFCS classification in infants is less precise than classification over time in older children. In conclusion, children can be classified by the GMFCS early on, but there is a need for reclassification at age 2 or older as more clinical information becomes available.