We adopted a species’ perspective for predicting extinction risk in a small, endemic, and strictly scansorial lizard (Urosaurus nigricaudus), in an old (∼60 year) and highly fragmented (8% habitat remaining) agricultural landscape from the Sonoran Desert, Mexico. We genotyped 10 microsatellite loci in 280 individuals from 11 populations in fragmented and continuous habitat. Individual dispersal was restricted to less than 400 m, according to analyses of spatial autocorrelation and spatially explicit Bayesian assignment methods. Within this scale, continuous areas and narrow washes with native vegetation allowed high levels of gene flow over tens of kilometers. In the absence of the native vegetation, cleared areas and highways were identified as partial barriers. In contrast, outside the scale of dispersal, cleared areas behaved as complete barriers, and surveys corroborated the species went extinct after a few decades in all small (less than 45 ha), isolated habitat fragments. No evidence for significant loss of genetic diversity was found, but results suggested fragmentation increased the spatial scale of movements, relatedness, genetic structure, and potentially affected sex-biased dispersal. A plausible threshold of individual dispersal predicted only 23% of all fragments in the landscape were linked with migration from continuous habitat, while complete barriers isolated the majority of fragments. Our study suggested limited dispersal, coupled with an inability to use a homogeneous and hostile matrix without vegetation and shade, could result in frequent time-delayed extinctions of small ectotherms in highly fragmented desert landscapes, particularly considering an increase in the risk of overheating and a decrease in dispersal potential induced by global warming.