Aims. This paper presents a detailed comparison between high-redshift observations from the VIMOS-VLT Deep Survey (VVDS) and predictions from the Munich semi-analytical model of galaxy formation. In particular, we focus this analysis on the magnitude, redshift, and colour distributions of galaxies, as well as their clustering properties. Methods. We constructed 100 quasi-independent mock catalogues, using the output of the semi-analytical model presented in De Lucia & Blaizot (2007, MNRAS, 375, 2). We then applied the same observational selection function of the VVDS-Deep survey, so as to carry out a fair comparison between models and observations. Results. We find that the semi-analytical model reproduces well the magnitude counts in the optical bands. It tends, however, to overpredict the abundance of faint red galaxies, in particular in the i′ and z′ bands. Model galaxies exhibit a colour bimodality that is only in qualitative agreement with the data. In particular, we find that the model tends to overpredict the number of red galaxies at low redshift and of blue galaxies at all redshifts probed by VVDS-Deep observations, although a large fraction of the bluest observed galaxies is absent from the model. In addition, the model overpredicts by about 14 per cent the number of galaxies observed at 0.2 < z < 1 with IAB < 24. When comparing the galaxy clustering properties, we find that model galaxies are more strongly clustered than observed ones at all redshift from z = 0.2 to z = 2, with the difference being less significant above z ≃ 1. When splitting the samples into red and blue galaxies, we find that the observed clustering of blue galaxies is well reproduced by the model, while red model galaxies are much more clustered than observed ones, being principally responsible for the strong global clustering found in the model. Conclusions. Our results show that the discrepancies between Munich semi-analytical model predictions and VVDS-Deep observations, particularly in the galaxy colour distribution and clustering, can be explained to a large extend by an overabundance of satellite galaxies, mostly located in the red peak of the colour bimodality predicted by the model.