1.0-2.5um spectroscopy of faint sources should use the "Low Background" read mode. Keep the exposure time longer than 44 sec and shorter than a few minutes. Plan to discard the first spectral image of a sequence, because of dark current instability when the background changes (between acquisition imaging and science spectroscopy). Bright star observations are much shorter and should use the "Medium Background" read mode.

Spectroscopy in the L band can be done with exposures of 1-3 seconds - or longer, depending on the wavelengths of interest - and thus in the "Medium Background" read mode (the minimum recommended exposure time is 2.7 sec for this mode). However, for higher efficiency or for bright standards it should be done in the "High Background" mode (minimum recommended exposure time is 0.9 sec). Except for bright sources, many exposures should be coadded between nods (so that the total integration time between nods is 30-60 seconds).

Spectroscopy in the M band must be done in the "High Background" read mode. The 768x768 subarray should be used rather than the full array; it will result in greater efficiency (because of the shorter read time), and no spectral information is lost, because the excluded portions of the array are outside the bandpass of the blocking filter. Use of the 512x1024 subarray does not gain any improvement in readout time and is equivalent to using the full array. As at L multiple coadds should be used for faint targets. At f/6 flats can be taken with the 768x768 subarray but they saturate the full array.

Spectroscopic observations should nod the target along the slit for sky and detector defect removal. Smaller nods more accurately retain the target in the slit than do large nods. Although the slit is 50-90 arcsec long it is not necessary to use all of it on small targets. For a point source nods of 3 arcsec at f/6 are ample.

Bright targets (J<21.5, H<20.5, K<20.7) which give S/N>3 in 60s imaging integrations in good (IQ=70, CC=50) conditions can be acquired directly. The target will be centered in the slit by placing a near-IR image of the target at the center of the slit, before introducing the appropriate slit, blocking filter and grism. Fainter objects can be acquired via accurate user-supplied blind offsets from a nearby bright object. In this mode the bright reference (User1) star will be centered in the slit, and then the blind offsets will be applied to shift the science target into the slit. In the thermal IR (3-5um), because of the high background it is much easier (except for the most extreme red objects) to center a near-IR image of the target on the slit than an image in the thermal IR. Examples of both types of acquisitions are given in the NIRI OT library.

The current f/6 slits and grisms are designed for use at f/6. Generally speaking, higher spectral resolution cannot be achieved using the f/14 camera and the present set of grisms, and the wavelength coverage at f/14 will be severely reduced.

At f/32 (using adaptive optics) three grisms are available to cover the JHK bands although these provide somewhat narrower wavelength coverage than the f/32 grisms. In general, If NIFS is available it rather than NIRI should be used for JHK spectroscopy with AO.

Slitless spectroscopy may be useful for some near-IR programs. The background will be much higher, of course. Some observations may use narrow-band filters in conjunction with slitless grism spectroscopy to good advantage. This mode has not been tested yet.