Designs for Extremely Large Telescopes (ELTs) are quite well advanced, but the requirements of instruments have had limited impact. Since provision of a suitable environment for instruments is a critical aspect of all telescopes, we outline some well-known and some less-appreciated challenges of designing instruments for ELTs. A wide-field spectrometer (WFSPEC) with ~10 arcmin field-of-view, probably with AO correction of ground-layer seeing, illustrates the well-known difficulty of matching modern detector pixels to large (~0."3) images. The challenges of exploiting wide-field (1'-2' FOV) high-performance AO systems on ELTs are illustrated by a Multi-Object Multi-field Spectrometer and Imager (MOMSI), which provides imaging and integral-field spectroscopy, at near-diffraction-limited pixel scales, of targets in approximately 300 subfields each. This instrument, roughly equivalent to all the astronomical spectrometers yet built, extracts ~200 times less of the available information from the ELT's FOV than near-future instruments on 8-m class telescopes will do for their hosts. We emphasise the great size of such instruments (40-100 tonnes, 100-200 m3) and the need to accommodate this size in telescope plans. A third area of challenge is the exploitation of the potential capabilities of ELTs in the mid-IR, where they would offer powerful complements to JWST and ALMA; low-emissivity telescope designs and, possibly, cryogenic AO, may be needed. Finally, we outline the potential challenges of correcting atmospheric dispersion effects.