The emission spectrum of carbon dioxide has been studied by the electron beam excitation method used by Smyth and Arnott. The whole range of the spectrum from 6500 to 1400 has been examined and only the bands reported by Fox, Duffendack and Barker in the region from 2700 to 5000 have been observed. A particular effort was made to get in emission the extreme ultraviolet bands reported by Leifson in absorption but it was unsuccessful. The "second negative bands of carbon" are shown to be probably identical with parts of the Fox, Duffendack and Barker system. This system of some two hundred bands is partially analyzed into fifteen progressions of band heads given by the formula $\ensuremath{\nu}=26,271\ensuremath{-}{\ensuremath{\nu}}_{x}+1136.85v\ensuremath{-}1.85{v}^{2}$ where ${\ensuremath{\nu}}_{x}$ takes on a particular value for each progression and $v=0,1\ensuremath{\cdots}10$. About one hundred bands fit into this arrangement. This empirical analysis is shown to be consistent with a scheme of energy levels involving two normal modes of vibration (${{\ensuremath{\nu}}_{1}}^{\ensuremath{'}}$ and ${{\ensuremath{\nu}}_{2}}^{\ensuremath{'}}$) in the upper electronic state and two (${{\ensuremath{\nu}}_{1}}^{\ensuremath{'}\ensuremath{'}}$ and ${{\ensuremath{\nu}}_{3}}^{\ensuremath{'}\ensuremath{'}}$) in the lower electronic state. The $2{{\ensuremath{\nu}}_{2}}^{\ensuremath{'}}$ state is split into two separated by 166 ${\mathrm{cm}}^{\ensuremath{-}1}$; this and the values of the frequencies are consistent with Dennison's and Barker and Martin's analysis of the infrared spectrum but since the spectrum here analyzed presumably does not involve the normal state the frequencies are numerically different.