We use a large N-body simulation to examine the detectability of H i in emission at redshift z≃ 1, and the constraints imposed by current observations on the neutral hydrogen mass function of galaxies at this epoch. We consider three different models for populating dark matter haloes with H i, designed to encompass uncertainties at this redshift. These models are consistent with recent observations of the detection of H i in emission at z≃ 0.8. Whilst detection of 21-cm emission from individual haloes requires extremely long integrations with existing radio interferometers, such as the Giant Meter Radio Telescope (GMRT), we show that the stacked 21-cm signal from a large number of haloes can be easily detected. However, the stacking procedure requires accurate redshifts of galaxies. We show that radio observations of the field of the Deep Extragalactic Evolutionary Probe 2 (DEEP2) spectroscopic galaxy redshift survey should allow detection of the H i mass function at the 5–12σ level in the mass range 1011.4 h−1 M⊙≤Mhalo≤ 1012.5 h−1 M⊙, with a moderate amount of observation time. Assuming a larger noise level that corresponds to an upper bound for the expected noise for the GMRT, the detection significance for the H i mass function is still at the 1.7–3σ level. We find that optically undetected satellite galaxies enhance the H i emission profile of the parent halo, leading to broader wings as well as a higher peak signal in the stacked profile of a large number of haloes. We show that it is in principle possible to discern the contribution of undetected satellites to the total H i signal, even though cosmic variance limitation make this challenging for some of our models.