We present simultaneous Planck, Swift, Fermi, and ground-based data for 105 blazars belonging to three samples with flux limits in the soft X-ray, hard X-ray, and gamma-ray bands. Our unique data set has allowed us to demonstrate that the selection method strongly influences the results, producing biases that cannot be ignored. Almost all the BL Lac objects have been detected by Fermi-LAT, whereas ~40% of the flat-spectrum radio quasars (FSRQs) in the radio, soft X-ray, and hard X-ray selected samples are still below the gamma-ray detection limit even after integrating 27 months of Fermi-LAT data. The radio to sub-mm spectral slope of blazars is quite flat up to ~70GHz, above which it steepens to <\alpha>~-0.65. BL Lacs have significantly flatter spectra than FSRQs at higher frequencies. The distribution of the rest-frame synchrotron peak frequency (\nupS) in the SED of FSRQs is the same in all the blazar samples with <\nupS>=10^13.1 Hz, while the mean inverse-Compton peak frequency, <\nupIC>, ranges from 10^21 to 10^22 Hz. The distributions of \nupS and of \nupIC of BL Lacs are much broader and are shifted to higher energies than those of FSRQs and strongly depend on the selection method. The Compton dominance of blazars ranges from ~0.2 to ~100, with only FSRQs reaching values >3. Its distribution is broad and depends strongly on the selection method, with gamma-ray selected blazars peaking at ~7 or more, and radio-selected blazars at values ~1, thus implying that the assumption that the blazar power is dominated by high-energy emission is a selection effect. Simple SSC models cannot explain the SEDs of most of the gamma-ray detected blazars in all samples. The SED of the blazars that were not detected by Fermi-LAT may instead be consistent with SSC emission. Our data challenge the correlation between bolometric luminosity and \nupS predicted by the blazar sequence.