In recent years, considerable effort has been made to estimate the level of selection that acts on non-coding DNA using extended versions of the McDonald-Kreitman test that compares ratios of polymorphism to divergence for neutral and putatively selected sites. This test however does not take into account possible local effects caused by linked selection. We performed computer simulations to investigate how loose linkage within the region of interest affects estimates of selection and proportion of sites fixed by positive selection when a true neutral region independent of sites under selection is used as a neutral standard. Local reduction in neutral polymorphism in the presence of deleterious and/or advantageous mutations is always observed even when the recombination rate is very high (Nr=0.1). As a result, the intensity of negative selection is underestimated while the intensity of positive selection and the proportion of sites fixed by positive selection are often overestimated. We also investigate conditions under which local linkage effects can generate false signatures of positive selection. Finally, we show that additional factors such as the use of weakly selected sites as a neutral standard and the genomic location of the region under study may also affect estimates of selection and proportion of sites fixed by positive selection.