We make use of a semi-analytic cosmological model that includes simple prescriptions for dust attenuation and emission to make predictions for the observable and physical properties of galaxies that may be detected by the recently launched HerschelSpace Observatory in deep fields such as Great Observatories Origins Deep Survey-Herschel. We compare our predictions for differential galaxy number counts in the Herschel Photodetector Array Camera and Spectrometer (PACS) (100 and 160 μm) and Spectral and Photometric Imaging Receiver (SPIRE) (250, 350 and 500 μm) bands with available observations. We find very good agreement with the counts in the PACS bands, for the overall counts and for galaxies binned by redshift at z < 2. At z > 2 our model underpredicts the number of bright galaxies in the PACS bands by a factor of 10. The agreement is much worse for all three SPIRE bands and becomes progressively worse with increasing wavelength. We discuss a number of possible reasons for these discrepancies and hypothesize that the effect of blending on the observational flux estimates is likely to be the dominant issue. We note that the PACS number counts are relatively robust to changes in the dust-emission templates, at least for the three sets of templates that we have tested, while the predicted SPIRE number counts are more template dependent especially at low redshift. We present quantitative predictions for the relationship between the observed PACS 160 and SPIRE 250 μm fluxes and physical quantities such as halo mass, stellar mass, cold gas mass, star formation rate and total infrared (IR) luminosity, at different redshifts. We also present predictions for the radial sizes of Herschel-selected discs at high redshift (z > 2) and find reasonable agreement with the available observations. Finally, we present quantitative predictions for the correlation between the PACS 160 μm flux and the probability that a galaxy has experienced a recent major or minor merger. Although our models predict a strong correlation between these quantities, such that more IR-luminous galaxies are more likely to be merger driven, we find that a significant fraction (more than half) of all high-redshift IR-luminous galaxies detected by Herschel are able to attain their high star formation rates without enhancement by a merger.