W. 2004. Balanced dispersal or source Á/sink Á/ do both models describe wood mice in farmed landscapes? Á/ Oikos 106: 536 Á/550.To test two models of how wood mouse Apodemus sylvaticus populations in different patches might interact, we estimated parameters from capture-mark-recapture data in four habitats (set-aside, crop, boundary and woodlot) at two sites in arable farmland. In the source Á/sink model, populations in 'source' patches have fitness/1, while in 'sink' patches fitness isB/1; dispersal is constrained so there is a net flow of individuals from sources to sinks. In the balanced dispersal model, patches may vary in quality and carrying capacity, but there are no sinks and no constraints on dispersal; fitness is equal across patches. Our results broadly support balanced dispersal. Patches close together were more likely to exchange individuals than those far apart, there was no evidence of density-dependent dispersal and little evidence for directionality of movement. Overall population growth rates were similar in all patches (range: 0.99 Á/1.05). Times when no animals were known to be alive occurred in some patches and could be attributed to stochastic events. On one crop patch, however, extinction followed high abundance, and coincided with poor recruitment, suggesting sink dynamics. Recruitment rates were highest in boundary and crop, but surprisingly, the proportion of individuals in reproductive condition, timing of breeding and pregnancy rates did not vary between habitats. Apparent survival rates were lowest in crop, although recapture rates were high. While female body size in December did not vary between habitats, males on setaside were smaller than elsewhere. In general, the balanced dispersal model best described arable wood mouse populations, but dynamics on one patch suggested an 'attractive' sink. Our results demonstrate that more than one model for dispersal dynamics between populations may apply within the same landscape, for the same species.