The Fire and Fire Surrogate study—a replicated, manipulative experiment—sought the most economically and ecologically efficient way to restore the nation's fire-maintained ecosystems. As part of this study, we conducted a 3-year mark–recapture study, comprising 105,000 trap-nights, to assess demographic responses of cotton mice (Peromyscus gossypinus) to Fire and Fire Surrogate treatments at the Gulf Coastal Plain site, where longleaf pine was the ecosystem to be restored. We compared competing models to evaluate restoration effects on variation in apparent survival and recruitment over time, space, and treatment, and incorporated measures of available source habitat for cotton mice with reverse-time modeling to infer immigration from outside the study area. The top-ranked survival model contained only variation over time, but the closely ranked 2nd and 3rd models included variation over space and treatment, respectively. The top 4 recruitment models all included effects for availability of source habitat and treatments. Burning appeared to degrade habitat quality for cotton mice, showing demographic characteristics of a sink, but treatments combining fire with thinning of trees or application of herbicide to the understory appeared to improve habitat quality, possibly creating sources. Bottomland hardwoods outside the study also acted as sources by providing immigrants to experimental units. Models suggested that population dynamics operated over multiple spatial scales. Treatments applied to 15-ha stands probably only caused local variation in vital rates within the larger population.