The Caucasian Grouse (Lyrurus mlokosiewiczi) is of conservation concern because it has been declining at least since the 1980s and has disappeared from parts of its range. We quantified landscape composition and configuration around Caucasian Grouse lek sites and random points within the Arasbaran Region in northeastern Azerbaijan, Iran at multiple scales (200-, 500-, and 1,000-m radii circles). We developed averaged logistic regression models based on Akaike Information Criterion weights for each scale. Habitat suitability models indicated good discrimination based on the receiver-operating characteristic (ROC) criteria (three averaged-models had ROC values >0.8), indicating landscape pattern can explain distribution of leks reasonably well. Unconditional, model-averaged estimates and 95% confidence intervals for beta estimates of variables included within buffer-width models suggest leks in the 200-m buffer width are in areas with lower standard deviation in patch area. Lek sites in the 500-m buffer-width models were predicted to be absent in landscapes where patches are more insular and have higher edge contrast. Lek sites were more likely to be present in landscapes having larger patches that were more irregular in shape and in habitats containing more patch types and greater proportion of forest and range cover types. The model-averaged estimates for the explanatory variables within the 1,000-m buffer width indicated lek sites were more likely to be present in landscapes that have high spatial heterogeneity as patches are more insular and contain the most forest and range cover types.