Current descriptive models of edge effects describe little more than the relationship between perimeter and area. A more realistic model, one that incorporates the additive nature of edge effects, is presented and tested. The total edge effect at a location within a patch is modelled as the sum of edge effects at points along the edges of the patch, weighted by the distances from the edge points to the location. At four sites in the central Amazon °80 km north of Manaus, Brazil, I measured vertical stratification of foliage in three primary forest habitats: (1) continuous forest close to linear edges, (2) 10—ha forest fragments, and (3) 1—ha forest fragments. Foliage thickness (foliage density along a vertical line) was scored in six height intervals (0—2, 2—5, 5—10, 10—20, 20—30, and 30—40 m). With decreasing distance from a linear edge, understory (0—5 m) foliage thickness increased and overstory (10—30 m) foliage thickness decreased, and the relationship between foliage thickness and distance to the edge agreed with that predicted by the model. Using parameter estimates obtained from the best “fit” of the model to these data from linear edges, understory and overstory thickness in 1— and 10—ha fragments was predicted. Foliage thickness varied extensively among fragments, but agreed quite closely with that predicted by the model, and the additive model performed better than a curvilinear non—additive one.