Large herbivores respond to seasonal changes in resource availability through habitat selection. Understanding variations in habitat choice is crucial for targeting conservation efforts, particularly for endangered, wide-ranging species, such as the Asian elephant. We assessed patterns and determinants of elephant habitat use during the dry season, a period of resource limitation in tropical deciduous forests, in the Western Ghats of Karnataka, India. We collected detection/non-detection data on elephant signs under an occupancy sampling framework, using spatially replicated surveys on foot along forest trails to estimate probabilities of habitat use by elephants. Each of our 97 sites (sampling units) was a grid cell of 11.75 km2 area. Data were analysed using an occupancy model, which estimated detection probabilities for signs, while explicitly addressing the potential spatial dependence between sign detections on adjacent replicates. Using covariates that are likely to influence resource use, we made ecological predictions about dry season habitat use by elephants across the study area of 1850 km2. The site-level probabilities of habitat use by elephants ranged from ψ^(SE^[ψ^])=0.04(0.15)to0.99(0.01). The estimated replicate level detection probability was p^t(SE^[p^t])=0.67(0.06). We found that distance to rivers was the best predictor of elephant habitat use, in dry season, demonstrating the overarching importance of riparian habitats in the landscape for the species. Artificial water holes established by wildlife managers do not appear to influence elephant habitat use, which is likely a result of abundant and near-uniform distribution of such water holes across the study area. The sign survey-based occupancy modelling approach provides a basis for reliable cost-effective assessment of spatial distribution and habitat use by elephants and other large herbivores. Such assessments are essential for effective conservation management of large herbivores.