Land-use change threatens a large number of tropical species (so-called ‘loser’ species), but a small subset of disturbance-adapted species may proliferate in human-modified landscapes (‘winner’ species). Identifying such loser and winner species is critically needed to improve conservation plans, but this task requires longitudinal studies that are extremely rare. We assessed this topic with small rodent assemblages in the Lacandona rainforest, a relatively new and highly dynamic agricultural frontier from southeastern Mexico. In particular, we measured the abundance of four rodent species in 12 forest sites during a 6 year period. We related changes in abundance to differences across time in landscape structure (i.e., percentage of forest cover, matrix contrast, number of forest patches, and forest edge density) surrounding each site. Total rodent abundance was almost two times higher in 2016 than in 2011, although abundances were generally low in all years. The abundance of Heteromys desmarestianus increased through time, mainly in forest sites with increasing matrix contrast. Oryzomys sp. also tended to increase in abundance, especially in sites with decreasing edge density. Sigmodon toltecus remained stable through time, but Peromyscus mexicanus tended to decrease in abundance, particularly in sites with decreasing edge density and increasing matrix contrast across time. Therefore, spatial variations in landscape structure lead to species-specific responses. If current deforestation rates persist, we predict a population decline of forest-specialist species (P. mexicanus), and an increase in generalist species (S. toltecus and Oryzomys sp.). Improving matrix quality is crucial for preventing the extinction of forest-specialist rodent species.