In this study, we sought to understand how landscape structure affects roe deer movements within their home-range in a heterogeneous and fragmented agricultural system of south-western France. We analysed the movements of 20 roe deer fitted with GPS collars which recorded their locations every 2–6 h over several months (mean = 9 months). Based on empirical observations and previous studies of roe deer habitat use, we hypothesised that roe deer should avoid buildings and roads, move preferentially along valley bottoms and through the more wooded areas of the landscape. To test these hypotheses we paired each observed movement step with 10 random ones. Using conditional logistic regression, we modelled a step selection function, which represents the probability of selecting a given step as a function of these landscape variables. The selectedmodel indicated thatmovements were influenced by all the tested landscape features, but not always in the predicted direction: our results suggested that roe deer tend to avoid buildings, roads, valley bottoms and possibly the more wooded areas (although the latter result should be interpreted with caution, as it may be influenced by a bias in the rate of GPS fix acquisition in woods). The distances to buildings and to roads were the most influential variables in the model, suggesting that the avoidance of potential sources of disturbancemay be a key factor in determining ranging behaviour of roe deer in human dominated landscapes.