High-resolution simulations of star-forming massive galactic discs have shown that clumps form with a characteristic baryonic mass in the range 107–108 M⊙, with a small tail exceeding 109 M⊙ produced by clump–clump mergers. This is in contrast with the observed kpc-size clumps with masses up to 1010 M⊙ in high-redshift star-forming galaxies. In this paper, we show that the comparison between simulated and observed star-forming clumps is hindered by limited observational spatial resolution and sensitivity. We post-process high-resolution hydrodynamical simulations of clumpy discs using accurate radiative transfer to model the effect of ionizing radiation from young stars and to compute H α emission maps. By comparing the intrinsic clump size and mass distributions with those inferred from convolving the H α maps with different Gaussian apertures, we mimic the typical resolution used in observations. We found that with 100 pc resolution, mock observations can recover the intrinsic clump radii and stellar masses, in agreement with those found by lensing observations. Instead, using a 1 kpc resolution smears out individual clumps, resulting in their apparent merging. This causes significant overestimations of the clump radii and therefore masses derived using methods that use their observed sizes. We show that limited sensitivity can also force observations to significantly overestimate the clump masses. We conclude that a significant fraction of giant clumps detected in the observations may result from artificially inflated radii and masses, and that ≈100 pc spatial resolution is required to capture correctly the physical characteristics of star-forming clumps if they are coherent structures produced by disc fragmentation.