In order to understand how trees replace themselves during the conversion from agricultural field back into eastern deciduous forest, I first used data sampled in ten old fields over the first 20 years after abandonment to generate six different models of tree replacement. Each of the six models was then used to generate abundance predictions for six common tree taxa (species or genera) from year 21 to 30. Finally each model tested those predictions against the actual data sampled during those same years (21–30) focusing on the predicted rank abundances of the six test tree taxa using Spearman's rank correlation coefficients. I found that: 1) models generally predicted tree abundances and tree replacements well, with the exception of Juniperus virginiana, 2) although some matrix models became constant and some patch models became zero over time, most of the time matrix models had a higher abundance prediction than the patch models for the same species, and 3) among the matrix models, model 3 (which used cover values) had the highest values—especially for the wind-dispersed species—and among patch models, model 4 (which used the number of plots) was often as great or greater than models 5 or 6 (which used stem counts and cover values, respectively)—especially for the bird-dispersed species. Taken together the results suggested that models using cover were closer to the data sampled in the fields compared to the other models within their type (matrix or patch), and that patch models were better estimators of stem densities than matrix models, perhaps due to the inherent dampening properties of matrixes.