Several observational works have attempted to isolate the effects of galaxy interactions by comparing galaxies in pairs with isolated galaxies. However, different authors have proposed different ways to build these so-called control samples (CS). By using mock galaxy catalogues of the SDSS-DR4 built up from the Millennium Simulation, we explore how the way of building a CS might introduce biases which could affect the interpretation of results. We make use of the fact that the physics of interactions is not included in the semianalytic model, to infer that any difference between the mock control and pair samples can be ascribed to selection biases. Thus, we find that galaxies in pairs artificially tend to be older and more bulge-dominated, and to have less cold gas and different metallicities than their isolated counterparts. Also because of a biased selection, galaxies in pairs tend to live in higher density environments, and in haloes of larger masses. We find that imposing constraints on redshift, stellar masses and local densities diminishes the selection biases by ~70%. Based on these findings, we suggest observers how to build an unique and unbiased CS in order to reveal the effect of galaxy interactions.