It can be difficult to further scientific understanding of rare or endangered species that live in inaccessible habitat using traditional methods, such as probabilistic modeling based on field data collection. Remote sensing (RS) can be an important source of information for the study of these animals. A key advantage of RS is its ability to provide information over an animal's complete range, but models incorporating RS data are limited by RS's ability to detect important habitat features. In this study, we focus on the rare, poorly-understood mountain bongo antelope (Tragelaphus euryceros isaaci) which survives in the wild in isolated pockets of montane forest in Kenya. We hypothesize that mountain bongo habitat is multi-scaled. We analyzed field and RS data (derived from SPOT, ASTER, and MODIS) ranging in scale from 0.02–85.93 ha to test our hypothesis. Important microhabitat features were identified through logistic regression models of vegetation structure data collected in plots (0.04 ha) of bongo presence (n = 36) and absence (n = 90). Models were selected using an information theoretic approach. We analyzed the correlations between microhabitat (four canopy and four understorey structure measures) and RS variables derived using spectral mixture (SMA) and texture analysis; most ASTER and SPOT variables were significantly related with canopy structure variables (max|r| = 0.56), but correlations between understorey structure and all but two RS variables were insignificant. Further logistic regression modeling showed that combining field microhabitat (primarily understorey structure variables) and larger-scaled RS measures (ASTER spectral mixture analysis variables aggregated to 450 m (20.25 ha)) provided superior models of bongo habitat selection than those based on field or RS data only. The results demonstrate that: 1) forest canopy characteristics at scales of ~ 20 ha and understorey structural conditions at the micro-scale of 0.04 ha were the most important features influencing bongo habitat selection; 2) models for predicting bongo habitat distribution must incorporate both micro- and macro-habitat variables; 3) optical RS data may characterize important micro-scale canopy variables with reasonable accuracy, but are ineffective for detecting understorey features (unless alternative techniques such as forest structural indices can be successfully applied); 4) RS and field data are both essential for understanding bongo habitat selection. The technique employed here for understanding this rare antelope's habitat selection may also be applied in studies of other large herbivores.