We discuss the possibility of performing blind surveys to detect large-scale features of the Universe using 21-cm emission. Using instruments with ~5-10 arcmin resolution currently in the planning stage, it should be possible to detect virialized galaxy clusters at intermediate redshifts using the combined emission from their constituent galaxies, as well as less overdense structures, such as protoclusters and the `cosmic web', at higher redshifts. Using semi-analytic methods, we compute the number of virialized objects and those at turnaround which might be detected by such surveys. We find that there is a surprisingly large number of objects even using small (~5 per cent) bandwidths and elaborate on some issues pertinent to optimizing the design of the instrument and the survey strategy. The main uncertainty is the fraction of neutral gas relative to the total dark matter within the object. We discuss this issue in the context of the observations which are currently available.