The spectrum of the flame of carbon monoxide burning in air and in oxygen at reduced pressure has been photographed on plates of high contrast which display the band spectrum clearly above the continuous background. Greater detail has been obtained than has been recorded previously and new measurements are given. The structure of the spectrum has been studied systematically. It is shown that the bands occur in pairs with a separation of about 60 cm.
-1
, this separation being due probably to the rotational structure. Various wave-number differences are found to occur frequently, and many of the strong bands are arranged in arrays using intervals of 565 and 2065 cm.
-1
. The possible origin of the spectrum is discussed. The choice of emitter is limited to a polyatomic oxide of carbon, of which carbon dioxide is the most likely. The spectrum of the suboxide C
3
O
2
shows some resemblance to the flame bands, but this molecule is improbable as the emitter on other grounds. A peroxide C0
3
is also a possibility, but no evidence for the presence of this has been obtained from experiments on the slow combustion of carbon monoxide. Carbon dioxide in gaseous or liquid form is transparent through the visible and quartz ultra-violet, and the flame bands are not obtained from CO
2
in discharge tubes. Comparison with the Schumann-Runge bands of oxygen shows that it is possible that the flame bands may form part of the absorption band system of CO
2
which is known to exist below 1700 A if there is a big change in shape or size of the molecule in the two electronic states. The electronic energy levels of CO
2
are discussed. Since normal CO
2
is not built up from normal CO and oxygen, an electronic rearrangement of the CO
2
must occur after the combustion process. Mulliken has suggested that the molecule in the first excited electronic state, corresponding to absorption below 1700 A, may have a triangular form. The frequencies obtained from the flame bands are compared with the infra-red frequencies of CO
2
. The 565 interval may be identified with the transverse vibration
v
2
, indicating that the excited electronic state is probably triangular in shape. The 2065 interval cannot, however, be identified with the asymmetric vibration
v
3
with any certainty. If the excited electronic state of CO
2
is triangular, then molecules formed during the combustion by transitions from this level to the ground state may be “vibrationally activated”. This is probably the reason for many of the peculiarities of the combustion of carbon monoxide.
Algebraic conditions and the sparsity of spectrally arbitrary patterns
