This work quantifies the influence of the terrestrial atmosphere on the instrumental polarization (IP) of telescopes. Our numerical simulations show that seeing mostly induces a random crosstalk between Stokes parameters. The main contamination affects the linear polarization signals and is due to mixing with intensity. In general, circular polarization signals are less corrupted by atmospheric effects. We have analyzed the IP considering the imaging mirrors and the entrance window of a particular instrument, namely the Large Earth-based Solar Telescope (LEST). Crosstalks up to 2 x 10-2 (window) and up to 2 x 10-3 (mirrors) occur in short-exposure images. This level of IP is largely reduced by increasing the exposure time or the pixel size. Means for extrapolating the IP of LEST to other instruments are provided. Using them, the dependence of the seeing-induced IP on wavelength, telescope size etc. is briefly examined.