High tides force shorebirds from intertidal feeding areas to sites known as roosts. We investigated the roost selection of great knots, Calidris tenuirostris, and red knots, Calidris canutus, on a tropical coastline in northwestern Australia, assessing several roost attributes and recording the frequency of use of each site through automatic radiotelemetry. To model roost choice we used two approaches: (1) conditional logistic regression models that assumed roost selection to be a trade-off based on a probabilistic assessment of several environmental characteristics; and (2) bounds-based models that assumed that birds selected the nearest roost site to their feeding grounds, provided that threshold values for certain environmental characteristics were met. Bounds-based models were more effective, and we suggest that they offer a closer approach to real roost choice mechanisms. By day, roost choice was affected by distance from the feeding area and microclimate; birds selected nearby roosts where they could stand on cool, wet substrates. Different roost selection criteria were used at night when birds chose safer, but more distant, roosts. Models that assumed that roost choice was influenced by recent experience of roost sites performed better than models that assumed constant assessment of roost quality. This effect was significant only at night, suggesting that memory was used more when information on roost quality was limited. Evidence that roost availability may influence selection of foraging areas is also presented. Our results suggest that shorebirds select roosts by using simple mechanisms, making roost choice models a potentially valuable tool in conservation planning.