Achieving the demanding sensitivity and bandwidth, envisaged for third-generation gravitational wave (GW) observatories, is extremely challenging with a single broadband interferometer. Very high optical powers (megawatts) are required to reduce the quantum noise contribution at high frequencies, while the interferometer mirrors have to be cooled to cryogenic temperatures in order to reduce thermal noise sources at low frequencies. To resolve this potential conflict of cryogenic test masses with high thermal load, we present a conceptual design for a 2-band xylophone configuration for a third-generation GW observatory, composed of a high-power, high-frequency interferometer and a cryogenic low-power, low-frequency instrument. Featuring inspiral ranges of 3200 Mpc and 38 000 Mpc for binary neutron stars and binary black holes coalesences, respectively, we find that the potential sensitivity of xylophone configurations can be significantly wider and better than what is possible in a single broadband interferometer. © 2010 IOP Publishing Ltd.